The Future of
Computer Trading
in Financial Markets
An International Perspective
FINAL PROJECT REPORT
This Report should be cited as:
Foresight: The Future of Computer Trading in Financial Markets (2012)
Final Project Report
The Government Ofce for Science, London
The Future of
Computer Trading
in Financial Markets
An International Perspective
This Report is intended for:
Policy makers, legislators, regulators and a wide range of professionals and researchers whose interest
relate to computer trading within nancial markets. This Report focuses on computer trading from an
international perspective, and is not limited to one particular market.
Foreword
Well functioning nancial markets are vital for everyone. They
support businesses and growth across the world. They provide
important services for investors, from large pension funds to the
smallest investors. And they can even affect the long-term security
of entire countries.
Financial markets are evolving ever faster through interacting
forces such as globalisation, changes in geopolitics, competition,
evolving regulation and demographic shifts. However, the
development of new technology is arguably driving the fastest
changes. Technological developments are undoubtedly fuelling
many new products and services, and are contributing to the
dynamism of nancial markets. In particular, high frequency
computer-based trading (HFT) has grown in recent years to
represent about 30% of equity trading in the UK and possible
over 60% in the USA.
HFT has many proponents. Its roll-out is contributing to fundamental shifts in market structures
being seen across the world and, in turn, these are signicantly affecting the fortunes of many market
participants. But the relentless rise of HFT and algorithmic trading (AT) has also attracted considerable
controversy and opposition. Some question the added value it brings to markets and, indeed, whether
it constitutes a drag on market efciency. Crucially, some also believe that it may be playing an
increasing role in driving instabilities in particular markets. This is of concern to all nancial markets,
irrespective of their use of HFT, since increasing globalisation means that such instabilities could
potentially spread through contagion. It has also been suggested that HFT may have signicant negative
implications relating to market abuse. For these reasons, it is unsurprising that HFT is now attracting
the urgent attention of policy makers and regulators across the world.
This international Foresight Project was commissioned to address two critical challenges. First, the pace
of technological change, coupled with the ever-increasing complexity of nancial trading and markets,
makes it difcult to fully understand the present effect of HFT/AT on nancial markets, let alone to
develop policies and regulatory interventions that are robust to developments over the next decade.
Second, there is a relative paucity of evidence and analysis to inform new regulations, not least because
of the time lag between rapid technological developments and research into their effects. This latter
point is of particular concern, since good regulation clearly needs to be founded on good evidence and
sound analysis.
Therefore, the key aim of this Project has been to assemble and analyse the available evidence concerning
the effect of HFT on nancial markets. Looking back through recent years and out to 2022, it has taken
an independent scientic view. The intention has been to provide advice to policy makers. Over 150
leading academics from more than 20 countries have been involved in the work which has been informed
by over 50 commissioned papers, which have been subject to independent peer review.
The key message is mixed. The Project has found that some of the commonly held negative
perceptions surrounding HFT are not supported by the available evidence and, indeed, that HFT
may have modestly improved the functioning of markets in some respects. However, it is believed
that policy makers are justied in being concerned about the possible effects of HFT on instability in
nancial markets. Therefore, this Report provides clear advice on what regulatory measures might be
most effective in addressing those concerns in the shorter term, while preserving any benets that
HFT/AT may bring. It also advises what further actions should be undertaken to inform policies in the
longer term, particularly in view of outstanding uncertainties. In conclusion, it is my pleasure to make
this Report and all of its supporting evidence and analysis freely available. It is my hope that it will
provide valuable insights into this crucial issue.
Professor Sir John Beddington CMG, FRS
Chief Scientic Adviser to HM Government and
Head of the Government Ofce for Science
Lead expert group overseeing
the Project:
Dame Clara Furse (Chair) Non-executive Director, Legal & General plc, Amadeus IT
Holding SA, Nomura Holdings Inc., Chairman, Nomura Bank
International, Non-executive Director, Department for Work
and Pensions and Senior Adviser, Chatham House.
Professor Philip Bond
Professor Dave Cliff
Professor Charles Goodhart CBE, FBA
Visiting Professor of Engineering Mathematics and Computer
Science at the University of Bristol and Visiting Fellow at the
Oxford Centre for Industrial and Applied Mathematics.
Professor of Computer Science at the University of Bristol.
Professor (Emeritus) of Banking and Finance at the London
School of Economics.
Kevin Houstoun Chairman of Rapid Addition and co-Chair of the Global
Technical Committee, FIX Protocol Limited.
Professor Oliver Linton FBA
Dr Jean-Pierre Zigrand
Chair of Political Economy at the University of Cambridge.
Reader in Finance at the London School of Economics.
Foresight would like to thank Dr Sylvain Friederich, University of Bristol, Professor Maureen O’Hara,
Cornell University and Professor Richard Payne, Cass Business School, City University, London for their
involvement in drafting parts of this Report.
Foresight would also like to thank Andy Haldane, Executive Director for Financial Stability at the Bank
of England, for his contribution in the early stages of the Project.
Foresight Project team:
Professor Sandy Thomas Head of Foresight
Derek Flynn Deputy Head of Foresight
Lucas Pedace Project Leader
Alexander Burgerman Project Manager
Gary Cook Project Manager
Christopher Grifn Project Manager
Anne Hollowday Project Manager
Jorge Lazaro Project Manager
Luke Ryder Project Manager
Piers Davenport Project Co-ordinator
Martin Ford Project Co-ordinator
Yasmin Hossain Project Researcher
Zubin Siganporia Project Researcher
Isabel Hacche Intern
Arun Karnad Intern
Louise Pakseresht Intern
Jennifer Towers Intern
For further information about the Project please visit:
http://www.bis.gov.uk/foresight
Contents
Executive Summary 9
1: Introduction 19
2: The impact of technology developments 27
3: The impact of computer-based trading on liquidity, 41
price efciency/discovery and transaction costs
4: Financial stability and computer-based trading 61
5: Market abuse and computer-based trading 87
6: Economic impact assessments on policy measures 99
6.1 Notication of algorithms 101
6.2 Circuit breakers 102
6.3 Minimum tick sizes 106
6.4 Obligations for market makers 108
6.5 Minimum resting times 111
6.6 Order-to-execution ratios 113
6.7 Maker-taker pricing 115
6.8 Central limit order book 117
6.9 Internalisation 118
6.10 Order priority rules 120
6.11 Periodic call auctions 122
6.12 Key interactions 123
7: Computers and complexity 131
8: Conclusions and future options 139
Annex A: Acknowledgements 147
Annex B: References 156
165
Annex D: Project reports and papers
Annex C: Glossary of terms and acronyms
172
Annex E: Possible future scenarios for computer-based trading in nancial markets 174
7
Executive summary
A key message: despite commonly held negative perceptions, the available evidence indicates that high
frequency trading (HFT) and algorithmic trading (AT) may have several benecial effects on markets.
However, HFT/AT may cause instabilities in nancial markets in specic circumstances. This Project has
shown that carefully chosen regulatory measures can help to address concerns in the shorter term.
However, further work is needed to inform policies in the longer term, particularly in view of likely
uncertainties and lack of data. This will be vital to support evidence-based regulation in this controversial
and rapidly evolving eld.
1 The aims and ambitions of the Project
The Project’s two aims are:
• to determine how computer-based trading (CBT) in nancial markets across the world could evolve
over the next ten years, identifying potential risks and opportunities that this could present, notably
in terms of nancial stability
1
but also in terms of other market outcomes, such as volatility, liquidity,
price efciency and price discovery;
• to draw upon the available science and other evidence to provide advice to policy makers, regulators
and legislators on the options for addressing present and future risks while realising potential benets.
An independent analysis and an international academic perspective:
The analysis provides an independent view and is based upon the latest science and evidence. As such,
it does not constitute the views or policy of the UK or any other government.
Over 150 leading academics and experts from more than 20 countries have been involved in the work
which has been informed by over 50 commissioned scientic papers, which have been independently
peer reviewed. A further 350 stakeholders from across the world also provided advice on the key
issues to consider
2
.
2 Why the Project was undertaken
Well functioning nancial markets are vital for the growth of economies, the prosperity and well-being of
individuals, and can even affect the security of entire countries. Markets are evolving rapidly in a difcult
environment, characterised by converging and interacting macro- and microeconomic forces, such as
globalisation, changes in geopolitics, competition, evolving regulation and demographic shifts. However, the
development and application of new technology is arguably causing the most rapid changes in nancial
markets. In particular, HFT and AT in nancial markets have attracted considerable controversy relating to
their possible benets and risks.
While HFT and AT have many proponents, others question the added value they bring to markets, and
indeed whether they constitute a drag on market efciency. Crucially, some believe they may be playing
an increasingly signicant role in driving instabilities in particular markets. There have been suggestions
that HFT and AT may have signicant negative implications relating to market abuse. For these reasons,
and in view of the vital importance of nancial markets, both HFT and AT are now attracting the
urgent attention of policy makers and regulators across the world.
1 A list of denitions used in this Executive Summary can be found in Annex C of the Project’s Final Report.
2 A list of individuals who have been involved can be found in Annex A of the Project’s Final Report.
9
The Future of Computer Trading in Financial Markets
Two challenges for regulators:
Effective regulation must be founded on robust evidence and sound analysis. However, this Project addresses
two particular challenges currently faced by regulators:
• Rapid developments and applications of new technology, coupled with ever-increasing complexity of
nancial trading and markets make it difcult to fully understand the present effects of HFT and AT on
nancial markets and even more difcult to develop policies and regulatory interventions which will be
robust to developments over the next decade.
• There is a relative lack of evidence and analysis to inform the development of new regulations, not least
because of the time lag between rapid technological developments and research into their effects, and the
lack of available, comprehensive and consistent data.
These two challenges raise important concerns about the level of resources available to regulators in
addressing present and future issues. Setting the right level of resources is a matter for politicians.
However, unlocking the skills and resources of the wider international academic community could also
help. Here, a drive towards making better data available for analysis should be a key objective and the
experience of this Project suggests that political impetus could be important in achieving that quickly.
It makes sense for the various parties involved in nancial markets to be brought together in framing further
analytical work, in order to promote wide agreement to the eventual results. Everyone will benet from
further research that addresses areas of controversy, as these can cloud effective and proportionate
policy development, and can result in sub-optimal business decisions.
3 Technology as a key driver of innovation and change in financial markets
3
The relentless development and deployment of new technologies will continue to have profound effects on
markets at many levels. They will directly affect developments in HFT/AT and continue to fuel innovation in
the development of new market services. And they will also help to drive changes in market structure.
New technologies are creating new capabilities that no human trader could ever offer, such as
assimilating and integrating vast quantities of data and making multiple accurate trading decisions
on split-second time-scales. Ever more sophisticated techniques for analysing news are also being
developed and modern automated trading systems can increasingly learn from monitoring sequences
of events in the market. HFT/AT is likely to become more deeply reliant on such technologies.
Future developments with important implications:
• There will be increasing availability of substantially cheaper computing power, particularly through cloud
computing: those who embrace this technology will benet from faster and more intelligent trading
systems in particular.
• Special purpose silicon chips will gain ground from conventional computers: the increased speed will
provide an important competitive edge through better and faster simulation and analysis, and within
transaction systems.
• Computer-designed and computer-optimised robot traders could become more prevalent: in time, they
could replace algorithms designed and rened by people, posing new challenges for understanding
their effects on nancial markets and for their regulation.
• Opportunities will continue to open up for small and medium-sized rms offering ‘middleware’ technology
components, driving further changes in market structure: such components can be purchased and plugged
together to form trading systems which were previously the preserve of much larger institutions.
10
3 For a more detailed review of the evidence reported in this section, see Chapter 2 in the Project’s Final Report.
Executive Summary
4
Three key challenges arising from future technological developments:
• The extent to which different markets embrace new technology will critically affect their competitiveness
and therefore their position globally: The new technologies mean that major trading systems can
exist almost anywhere. Emerging economies may come to challenge the long-established historical
dominance of major European and US cities as global hubs for nancial markets if the former
capitalise faster on the technologies and the opportunities presented.
• The new technologies will continue to have profound implications for the workforce required to service
markets, both in terms of numbers employed in specic jobs, and the skills required: Machines can
increasingly undertake a range of jobs for less cost, with fewer errors and at much greater speed. As
a result, for example, the number of traders engaged in on-the-spot execution of orders has fallen
sharply in recent years, and is likely to continue to fall further in the future. However, the mix of
human and robot traders is likely to continue for some time, although this will be affected by other
important factors, such as future regulation.
• Markets are already ‘socio-technical’ systems, combining human and robot participants. Understanding
and managing these systems to prevent undesirable behaviour in both humans and robots will be key to
ensuring effective regulation: While this Report demonstrates that there has been some progress in
developing a better understanding of markets as socio-technical systems, greater effort is needed in
the longer term. This would involve an integrated approach combining social sciences, economics,
nance and computer science. As such, it has signicant implications for future research priorities.
The impact of computer-based trading on market quality: liquidity, price
efficiency/discovery and transaction costs
4
While the effect of CBT on market quality is controversial, the evidence available to this Project suggests
that CBT has several benecial effects on markets, notably:
• liquidity, as measured by bid-ask spreads and other metrics, has improved;
• transaction costs have fallen for both retail and institutional traders, mostly due to changes in trading
market structure, which are related closely to the development of HFT in particular;
• market prices have become more efcient, consistent with the hypothesis that CBT links markets and
thereby facilitates price discovery.
While the above improvements in market quality should not be overstated, they are important,
particularly since they counter the belief that HFT provides no useful function in nancial markets.
Nevertheless, there are concerns relating to market quality which are worthy of mention.
A particular concern:
While overall liquidity has improved, there appears to be greater potential for periodic illiquidity: The nature
of market making has changed, with high frequency traders now providing the bulk of such activity
in both futures and equities. However, unlike designated specialists, high frequency traders typically
operate with little capital, hold small inventory positions and have no obligations to provide liquidity
during periods of market stress. These factors, together with the ultra-fast speed of trading, create the
potential for periodic illiquidity. The US Flash Crash and other more recent smaller events illustrate this
increased potential for illiquidity.
A key message: regulatory changes in practices and policies will be needed to catch up to the new realities of
trading in asset markets. However, caution needs to be exercised to avoid undoing the advantages that HFT
has brought.
For a more detailed review of the evidence reported in this section, see Chapter 3 in the Project’s Final Report.
11
4
The Future of Computer Trading in Financial Markets
5 Financial stability and computer-based trading
5
The evidence available to this Project provides no direct evidence that computer-based HFT has increased
volatility in nancial markets. However, in specic circumstances CBT can lead to signicant instability. In
particular, self-reinforcing feedback loops, as well as a variety of informational features inherent in computer-
based markets, can amplify internal risks and lead to undesired interactions and outcomes. This can happen
even in the presence of well-intentioned management and control processes. Regulatory measures for
addressing potential instability are considered in Section 7 of this Executive Summary.
Three main mechanisms that may lead to instabilities and which involve CBT are:
• nonlinear sensitivities to change, where small changes can have very large effects, not least through
feedback loops;
• incomplete information in CBT environments where some agents in the market have more, or more
accurate, knowledge than others and where few events are common knowledge;
• internal ‘endogenous’ risks based on feedback loops within the system.
The feedback loops can be worsened by incomplete information and a lack of common knowledge.
A further cause of instability is social: a process known as ‘normalisation of deviance’, where
unexpected and risky events (such as extremely rapid crashes) come to be seen as increasingly normal,
until a disastrous failure occurs.
6 Market abuse
6
and computer-based trading
7
Economic research thus far, including the empirical studies commissioned by this Project, provides no direct
evidence that HFT has increased market abuse
8
. However, the evidence in the area remains tentative:
academic studies can only approximate market abuse as data of the quality and detail required to identify
abuse are simply not available to researchers.
This Project has commissioned three empirical studies that nd no direct evidence of a link between
HFT and market abuse. The main focus of these studies is not on the measurement of market abuse
during the continuous phase of trading, however. The Project has reviewed qualitative evidence on
perceived levels of manipulation from various sources including interviews with traders and investors,
the nancial press, UK and international regulatory reports, submissions to regulatory consultations
and large-scale surveys of market participants. A new survey of end users was also carried out by
the Project
9
.
This qualitative evidence consistently indicates high levels of concern. Claims of market manipulation using
HFT techniques are reported by institutional investors such as pension funds and mutual funds in different
countries. These claims are, in turn, widely relayed by the nancial press. Even if not backed by statistical
evidence, these perceptions need to be taken seriously by policy makers because, given that the true extent
of abuse is not precisely known, it is perception that is likely to determine the behaviour of liquidity suppliers.
High perceived levels of abuse may harm market liquidity and efciency for all classes of traders.
The qualitative evidence mentioned above is not easy to interpret unambiguously. It is consistent
with three different ‘scenarios’ that are not mutually exclusive:
• High frequency traders exploit their speed advantage to disadvantage other participants in
nancial terms.
• The growth of HFT has changed order ows in ways that facilitate market abuse by both
slow and fast agents (for example, by making ‘predatory trading’ easier).
5 For a more detailed review of the evidence reported in this section, see Chapter 4 in the Project’s Final Report.
6 Here the concern is with market abuse relating to manipulative behaviour, by which a market is temporarily distorted to one
party’s advantage. Abuse relating to insider trading is not considered here.
7 For a more detailed review of the evidence reported in this section, see Chapter 5 in the Project’s Final Report.
8 A list of the studies commissioned by the Project may be found in Annex D of the Project’s Final Report.
9 SR1 (Annex D refers).
12
Executive Summary
• Other market developments concomitant with the growth in HFT, but not necessarily brought
about by HFT growth, may have contributed to an increase in the perception or actual prevalence
of abuse. Fragmentation of liquidity across trading venues is an example.
Regulators and policy makers can inuence perceptions, even if denitive evidence on the extent of abuse
will not be available to settle the debate.
• Regulators can address the lack of condence that market participants have in their ability to detect
and prosecute abuse in HFT and fragmented markets. While this may require signicant investment
in regulatory activity, if progress is made, both the perception and reality of abuse will be reduced;
for abusers, even a perceived threat of being caught may be a powerful disincentive.
• More statistical evidence on the extent of HFT manipulation most often described by institutional
investors can be produced
10
. This will help to correct or conrm perceptions. It will also be important
in guiding regulatory action, as the three scenarios outlined above may have very different policy
implications.
Detecting evidence of market abuse from vast amounts of data from increasingly diverse trading platforms
will present a growing challenge for regulators.
To identify abuse, each national regulator will need access to international market data. Otherwise
the market abuser can hide by transacting simultaneously in several separately linked markets. In the
USA, the Ofce of Financial Research (OFR) has been commissioned by the Dodd-Frank Act to fund
a nancial data centre to collect, standardise and analyse such data. There may be case for a similar
initiative to be introduced in Europe.
7 Economic impact assessments of policy measures
11
A number of policies related to CBT are being considered by policy makers with the goals of improving
market efciency and reducing the risks associated with nancial instability. This Project has commissioned
a variety of studies to evaluate these policies, with a particular focus on their economic costs and benets.
12
The key conclusions are set out below.
Policy measures that could be effective:
• Circuit breakers: There is general support for these, particularly for those designed to limit periodic
illiquidity induced by temporary imbalances in limit order books. They are especially relevant to markets
operating at high speed. Different markets may nd different circuit breaker policies optimal, but in
times of overall market stress there is a need for coordination of circuit breakers across markets,
and this could be a mandate for regulatory involvement. New types of circuit breakers triggered by
ex-ante rather than ex-post trading may be particularly effective in dealing with periodic illiquidity.
However, further investigation is needed to establish how coordination could best be achieved in
the prevailing market structure.
• Tick size policy: This can have a large inuence on transactions costs, market depth and liquidity
provision. The current approach of allowing each European trading venue to choose its own
minimum tick size has merits, but this can result in a race to the bottom between venues. A uniform
policy applied across all European trading venues is unlikely to be optimal, but a coherent overall
minimum tick size policy applying to subsets of trading venues may be desirable. This coordinated
policy could be industry-based, such as the one agreed to recently by the Federation of European
Securities Exchanges (FESE) members.
10 ‘Quote stufng’ or order book ‘layering’ are obvious examples.
11 For a more detailed review of the evidence reported in this section, see Chapter 6 in the Project’s Final Report.
12 A list of the economic impact assessments commissioned by the Project can be found in Annex D of the
Project’s Final Report.
13
The Future of Computer Trading in Financial Markets
Policy measures that are likely to be problematic:
• Notication of algorithms: The implementation of this, even if feasible, would require excessive
costs for both rms and regulators. It is also doubtful that it would substantially reduce the risk of
market instability due to errant algorithmic behaviour.
• Imposing market maker obligations and minimum resting times on orders: The former issue runs
into complications arising from the nature of high frequency market making across markets, which
differs from traditional market making within markets. Requirements to post two-sided quotes may
restrict, rather than improve, liquidity provision. Similarly, minimum resting times, while conceptually
attractive, can impinge upon hedging strategies that operate by placing orders across markets and
expose liquidity providers to increased ‘pick-off risk’ due to the inability to cancel stale orders.
• Order-to-execution ratios: This would be a blunt instrument to reduce excessive message trafc
and cancellation rates. While it could potentially reduce undesirable manipulative strategies, it may
also curtail benecial strategies. There is not sufcient evidence at this point to ascertain these
effects, and so caution is warranted. Explicit fees charged by exchanges on excessive messaging, as
well as greater regulatory surveillance geared to detect manipulative trading practices may be more
desirable approaches to deal with these problems.
• Maker-taker pricing: The issue is complex and is related to other issues like order routing, priority
rules and best execution. Regulatory focus on these related areas seems a more promising way of
constraining any negative effects of maker-taker pricing than direct involvement in what is generally
viewed as an exchange’s business decision.
• The virtual central limit order book (CLOB): The introduction of competition between trading
venues brought about by Markets in Financial Instruments Directive (MiFID) has resulted in more
choices for investors and, in many cases, improved market quality, but it has also led to greater
complexity and risk. The virtual CLOB it has created is still evolving and improving, but its current
structure falls short of a single integrated market. This raises a number of issues for both individual
exchange and market behaviour.
• Constraining internalisation or, more generally, dark trading: Off-exchange trading can be mutually
benecial for all parties involved, especially where large orders are involved. However, the trend
away from pre-trade transparency cannot be continued indenitely without detrimental effects on
the public limit order book and price discovery. Constraining these activities within a range that does
not adversely affect price discovery but does allow for benecial trading is important but difcult.
Evidence gathered from European markets is too limited to give satisfactory guidance.
• Call auctions: These are an alternative trading mechanism that would eliminate most of the
advantage for speed present in modern electronic markets. They are widely used already in equity
markets at open and close and following a trading halt, although no major market uses them
exclusively. To impose call auctions as the only trading mechanism seems unrealistic and draconian.
There are serious coordination issues related to hedging strategies that would make this policy
undesirable.
Two words of caution: Whilst the above conclusions are consistent with the currently available
evidence, further empirical study is desirable for some of the policy measures in particular. It should
also be recognised that some of the above individual policy options interact with each other in
important ways. For example, the presence or absence of circuit breakers affects most other measures,
as does minimum tick sizes. Decisions on individual policies should not therefore be taken in isolation,
but should take account of such important interactions
13
.
14
13 See the Project’s Final Report (Chapter 6, Section 6.12) and also the supporting evidence papers which were commissioned
(Annex D of the Project’s Final Report refers).
Executive Summary
8 Computers and complexity
Over coming decades, the increasing use of computers and information technology in nancial systems is likely
to make them more, rather than less complex. Such complexity will reinforce information asymmetries and
cause principal/agent problems, which in turn will damage trust and make the nancial systems sub-optimal.
Constraining and reducing such complexity will be a key challenge for policy makers. Options include
requirements for trading platforms to publish information using an accurate, high resolution, synchronised
timestamp. Improved standardisation of connectivity to trading platforms could also be considered.
However, there is no ‘magic bullet’ to address this issue. Policy makers will need an integrated approach
based on improved understanding of nancial systems. This will need to be achieved through:
• Improved post-trade transparency: The challenge of ensuring adequate dissemination and storage
of trading data to enable market abuse to be identied provides an important example of where
improvements need to be considered.
• Analysis: Making sense of disclosed information and developing a better understanding of the
nancial system will be critical. This implies the need to harness the efforts of researchers
14
.
A further proposal that is sometimes made is that (various categories of) agents should only be allowed
to hold or issue instruments which have been approved by the authorities in advance. This contrasts
with the more common position that innovation should be allowed to ourish, but with the authorities
retaining the power to ban the uses of instruments where they consider evidence reveals undesirable
effects. The former stance, however, not only restricts innovation, but also such ofcial approval
may well have unintended consequences. Furthermore, the effectiveness of such ofcial approval
is debatable. Ofcials have no more, and probably less, skill in foreseeing how nancial instruments
will subsequently fare than credit rating agencies or market agents. Indeed, many, possibly all, of the
instruments now condemned in some quarters as having played a part in the recent global nancial
crisis would, at an earlier time, have probably been given ofcial approval.
A corrective step that could, and should, be taken is to simplify (electronic) nancial systems by the
application of greater standardisation, particularly in the form of accurate, high resolution, synchronised
timestamps. CBT, operating on many trading platforms, has led to a vast expansion of data, which are
often not standardised, nor easily accessible to third parties (for example, regulators and academics) for
analysis and research. The relevant authorities should consider following the US example and establish
a European Financial Data Centre to collect, standardise and analyse such data.
9 Conclusions – key priorities for action
15
While the effects CBT on nancial markets have been the topic of some controversy in recent years, analysis
of the available evidence has shown that CBT has led to benets to the operation of markets, notably relating
to liquidity, transaction costs and the efciency of market prices
16
. Against the background of ever greater
competition between markets, it is highly desirable that any new policies or market regulation preserve
these benets.
However, this Project has also highlighted legitimate concerns that merit the close attention of policy makers,
particularly relating to the possibility of instabilities occurring in certain circumstances, and also periodic
illiquidity
17
. In view of the critical importance of nancial markets for global growth and prosperity, the
following suggests priorities for action:
14 See Section 9 of this Executive Summary.
15 For a more detailed review of this section, see Chapter 8 in the Project’s Final Report.
16 See Section 4 of this Executive Summary.
17 See Section 5 of this Executive Summary.
15
The Future of Computer Trading in Financial Markets
A. Limiting possible future market disturbances:
A.1 European authorities
18
, working together, and with nancial practitioners and academics, should assess
(using evidence-based analysis) and introduce mechanisms for managing and modifying the potential adverse
side-effects of CBT and HFT. Section 7 of this Executive Summary sets out analysis of ten individual
policy options, and provides advice on which are supported most by the available evidence. It is also
important that such regulatory measures are considered together, not individually, in view of important
interactions which may exist between some of them.
A.2 Coordination of regulatory measures between markets is important and needs to take place at two levels:
• Regulatory constraints involving CBT in particular need to be introduced in a coordinated manner across
all markets where there are strong linkages.
• Regulatory measures for market control must also be undertaken in a systematic global fashion
to achieve in full the objectives they are directed at. A joint initiative from a European Ofce of
Financial Research and the US Ofce of Financial Research (OFR), with the involvement of other
international markets, could be one option for delivering such global coordination.
A.3 Legislators and regulators need to encourage good practice and behaviour in the nance and software
engineering industries. This clearly involves the need to discourage behaviour in which increasingly risky
situations are regarded as acceptable, particularly when failure does not appear as an immediate result
19
.
These recognise that nancial markets are essentially complex ‘socio-technical’ systems, in which both
humans and computers interact: the behaviour of computers should not be considered in isolation.
A.4 Standards should play a larger role. Legislators and regulators should consider implementing accurate,
high resolution, synchronised timestamps because this could act as a key enabling tool for analysis
of nancial markets. Clearly it could be useful to determine the extent to which common gateway
technology standards could enable regulators and customers to connect to multiple markets more
easily, making more effective market surveillance a possibility.
A.5 In the longer term, there is a strong case to learn lessons from other safety-critical industries, and to use
these to inform the effective management of systemic risk in nancial systems. For example, high-integrity
engineering practices developed in the aerospace industry could be adopted to help create safer
automated nancial systems.
B. Making surveillance of financial markets easier:
B.1 The development of software for automated forensic analysis of adverse/extreme market events would
provide valuable assistance for regulators engaged in surveillance of markets. This would help to address
the increasing difculty that people have in investigating events.
C. Improving understanding of the effects of CBT in both the shorter and longer term:
C.1 Unlocking the power of the research community has the potential to play a vital role in addressing the
considerable challenge of developing better evidence-based regulation relating to CBT risks and benets and
also market abuse in such a complex and fast-moving eld. It will also help to further address the present
controversy surrounding CBT. Suggested priorities include:
• Developing an ‘operational process map’: this would detail the processes, systems and interchanges
between market participants through the trade life cycle, and so help to identify areas of high
systemic risk and broken or failing processes.
• Making timely and detailed data across nancial markets easily available to academics, but recognising
the possible condentiality of such data.
16
18 While several of these potential actions for stakeholders are framed within the European context, they will also be relevant to
stakeholders in other parts of the world.
19 A term for behaviour which accepts increasingly risky situations in the absence of adverse effects is called ‘normalisation of
deviance’. See Section 5 of this Executive Summary.
Executive Summary
C.2 The above measures need to be undertaken on an integrated and coordinated international basis in
order to realise the greatest added value and efciency. One possible proposal would be to establish a
European Financial Data Centre.
In conclusion:
It is hoped that the analysis and arguments contained in the Foresight Final Project Report, together
with over 50 commissioned evidence papers which underpin it, will assist policy makers, regulators
and market practitioners in their current consideration of CBT. In this context, special thanks are
due to the 150 or so leading and independent experts from over 20 countries who have been involved
in this undertaking.
This Executive Summary and the underlying Project Report provide an independent view based on the best
available science and evidence. They do not constitute the views or policy of the UK or any other government.
17
1 Introduction
1.1 The aim of this Project
This Project has two principal aims. First, looking out to 2022, it seeks to determine how computer-
based trading in nancial markets could evolve and, by developing a robust understanding of its
effects, to identify potential risks and opportunities that this could present, notably in terms of nancial
stability
1
but also in terms of other market outcomes such as volatility
2
, liquidity
3
, price efciency and
price discovery
4
. Secondly, drawing upon the best available scientic and other evidence, the Project
aims to provide advice to policy makers, regulators and legislators on the options for addressing those
risks and opportunities.
1.2 Why the Project was commissioned
Computer-based trading (CBT)
5
has grown substantially in recent years, due to fast-paced
technological developments and their rapid uptake, particularly in equity markets. For example, possibly
30% of the UK’s equity trading volume is now generated through high frequency trading (HFT), while
in the USA this gure is possibly over 60%
6
. CBT is therefore already transforming the ways in which
nancial markets operate.
Inevitably, substantial changes to the functioning of nancial markets, actual or perceived, attract
considerable attention, not least because of their potential impact on market condence, the operation
of businesses and the health and growth of economies.
HFT attracts particular controversy. There has been a continuing debate about the extent to which
HFT improves or degrades the functioning of nancial markets, and also its inuence on market
volatility and the risk of instabilities. Indeed, such trading has been implicated by some as a contributory
factor in the Flash Crash of 6 May 2010 in which one trillion dollars temporarily evaporated from
US markets
7
. However, the wider debate on HFT and CBT has been hampered by the availability of
relatively little evidence, and a lack of analysis.
The controversy concerning the effect of CBT on nancial systems has the potential to grow for
several reasons. Relentless technological developments in both hardware and in trading algorithms,
together with other drivers of change
8
, continue to inuence the structure of markets, inevitably
creating winners and losers at all levels. They are also leading to increases in complexity as well as
new dynamics, making markets ever harder to understand and to regulate, particularly in view of the
rapid pace of change. These developments are fuelling an urgent need to gain a better understanding
of a range of issues, most notably concerning their effect on systemic risk of nancial instability and
its management. However, other important concerns relate to the effect of CBT on the efciency of
markets which has particularly divided opinion; the evolving potential for market abuse, especially its
1 Financial market stability refers to the lack of extreme movements in asset prices over short time periods. A glossary of terms
and a list of acronyms used in this Report can be found in Annex C.
2 Volatility is dened here as variability of an asset’s price over time, often measured in percentage terms.
3 Liquidity is dened here as the ability to buy or sell an asset without greatly affecting its price. The more liquid the market, the
smaller the price impact of sales or purchases. For a more detailed description see Section 3.3.1.
4 Price efciency and price discovery – pricing is efcient when an asset’s price reects the true underlying value of an asset;
price discovery refers to the market process whereby new information is impounded into asset prices.
5 See Box 1.1 for a denition of CBT.
6 Accurate estimates of the volume of high frequency trading are difcult to obtain, and in any case are contingent on the
precise denition used. For estimates, see Kaminska (2011), Kaminska (2009) and http://www.tradersmagazine.com/news/high-
frequency-trading-benets-105365-1.html?zkPrintable=true Accessed: 3 September 2012.
7 Chapter 4 reviews the available evidence for the inuence of HFT in the Flash Crash of 6 May 2010.
8 See Box 1.3 and Figure 1.1.
19
The Future of Computer Trading in Financial Markets
detection and regulation; and the relationship of CBT with dark pools and changing market institutions
more generally.
For these reasons, CBT is currently attracting the close attention of policy makers and regulators
worldwide. For example:
• In Europe, HFT has been placed near the top of the regulatory agenda, with a wide range of
measures on CBT being debated in the European Union (EU) Parliament and the EU Commission
within the Markets in Financial Instruments Directive (MiFID) II process. Indeed, some parties
have mooted measures that could have very far-reaching implications for markets, as they could
substantially constrain the future of CBT within Europe.
• In the USA, a number of measures are being proposed under the Dodd-Frank Act.
• A number of countries across the world, from Latin America to Asia, have adopted measures
related to CBT as exchanges turn electronic.
Against this background of rapid change and the urgent needs of regulators, the relative lack of
evidence and the prevalence of controversy over consensus is a major concern. Regulation that is not
informed by evidence and analysis risks making matters worse rather than better. It was therefore
decided to commission this Foresight Project to inform a broad audience of policy makers, regulators
and other stakeholders around the world.
Box 1.1: Definition of computer-based trading
Computer-based trading (CBT) refers to the trading system itself. Financial institutions use CBT
systems in a range of trading strategies, of which high frequency trading (HFT)
9
and algorithmic
trading (AT) are two types. However, the use of a CBT system by a nancial institution does not
necessarily mean that it is a user of one or other of these strategies.
A useful taxonomy of CBT systems identies four characteristics that can be used to classify CBT
systems
10
:
1) CBT systems may trade on an agency basis (i.e. attempting to get the best possible execution
of trades on behalf of clients) or a proprietary basis (i.e. trading using one’s own capital).
2) CBT systems may adopt liquidity-consuming (aggressive) or liquidity-supplying (passive)
trading styles.
3) CBT systems may engage in either uninformed or informed trading.
4) A CBT algorithm generates the trading strategy or only implements a decision taken by
another market participant
11
.
A more detailed denition may be found in Annex C.
20
9 For a denition of HFT please see Chapter 3.
10 DR5 (Annex D refers).
11 Please refer to Annex C for a comprehensive glossary and list of acronyms.
Introduction
1.3 A robust, international and independent approach
This Foresight Project has taken a broad approach, drawing upon a wide range of disciplines, including
economics, computer science, sociology and physics. In so doing, it has engaged with over 150 leading
independent academics from more than 20 countries
12
.
The Project has drawn upon a wider body of academic literature, but has also commissioned over 50
papers and studies, particularly where it was considered important to address gaps in the evidence
base, or to collate and assess existing studies. A full list of commissioned work is presented in Annex D:
all of this information is freely available from www.bis.gov.uk/foresight. Such work includes reviews of
drivers of change, economic impact assessments of regulatory measures and the results of workshops
held in Singapore, New York and London to gather the views of leading industry practitioners,
economists and technologists on future scenarios. A sociological study and a survey of end users
were also undertaken to understand the impact of computer trading on institutional investors. The
commissioned work has been peer reviewed internationally by independent reviewers (except for
certain workshop reports and a survey).
Throughout, the Project has been guided by a group of leading academics and senior industry
practitioners. It has also beneted from the advice of an international group of high-level stakeholders,
which has provided advice at critical stages of the Project. A full list of the 500 or so experts,
academics and stakeholders who have been involved in the Project is provided in Annex A.
Box 1.2: An independent view
While the Project has been managed by the UK Foresight programme under the direction of
Sir John Beddington, the UK Chief Scientic Adviser, its ndings are entirely independent of
the UK Government. As such, the ndings do not represent the views of the UK or any other
government, or the views of any of the organisations that have been involved in the work.
1.4 Project scope
The Project looks ten years into the future to take a long-term and strategic view of how CBT in
nancial markets might evolve, and how it might act within the context of other drivers of change, to
affect a range of market functions including: nancial stability, liquidity, price efciency and discovery,
transaction costs, technology and market abuse. While the future is inherently uncertain, major forces
driving change can be identied. A key driver of change is technology and this report explores how
technology and other drivers of change will interact to inuence the development and impact of CBT.
The possible consequences of a range of possible regulatory measures are also assessed.
Nine broad classes of drivers affecting nancial markets have been considered. These were identied
in three international industry workshops (held in London, Singapore and New York), and a workshop
of chief economists in London. These drivers are briey explained in Box 1.3 and are presented
diagrammatically in Figure 1.1. In particular, this gure shows that all of the drivers affect the key aspects
of market function, which in turn feed back to affect particular drivers
13
.
While the work has taken a global view of drivers of change and markets, it has paid particular attention
to the evolving use of CBT in Europe. However, much of the analysis will nevertheless be of interest to
policy makers and markets in other parts of the world.
12 Please refer to Annex A for a list of all the individuals involved in this Project.
13 Please refer to Annex E for discussion on how drivers of change could play out in alternative future scenarios.
21
The Future of Computer Trading in Financial Markets
The analysis in this Report focuses particularly on high frequency and algorithmic trading. Its aim is to
provide advice to policy makers who are taking decisions today on the regulation of CBT in markets
– so that those decisions are well founded and are more likely to be robust to future uncertainties.
However, in taking a ten year view, the Project also assesses what actions need to be implemented to
address systemic issues such as nancial instability, in the longer term. Some of these actions may, by
their nature, take longer for policy makers to agree and implement.
The Project has taken an evidence-based approach wherever possible. However, in such a fast-moving
eld, gaps in both understanding and evidence are inevitable and these are highlighted in the text
where appropriate. The most important gaps for informing regulation are identied in the concluding
chapter. The Project has also explored the current perceptions of leading institutions about the
impact of CBT in markets and importantly, it has evaluated the extent to which such perceptions are
supported by the available evidence.
Figure 1.1: Key drivers of change
Financial
stability
Liquidity
Price discovery/
efficiency
Transaction
costs
Market
integrity
Technology
Asset classes
Competition
Industry
workshops
Chief
Economists
workshop
Geopolitics
Regulation
Demographics
Global economic cycles
Loss/change of riskless
(reference) assets
Change in
(dis)intermediation
Market outcomes Key drivers of change
22
Introduction
Box 1.3: Important future drivers of change
The following provides a brief summary of the nine broad classes of drivers identied during three
international industry workshops (London, Singapore and New York), and a workshop of chief
economists in London
14
.
Regulation: Regulation will have an important and uncertain inuence on nancial markets, as both
a driver and a consequence of nancial market changes. As a driver, it may change the allocation
of investment across assets and exchanges, and across institutions and investment models or
strategies. Future regulation could be more or less coercive, informed by big data analytics,
sophisticated models, heavy- or light-touch. There could be fragmentation at the global level,
possibly linked to resurgent protectionism. Demand for regulation will tend to have an inverse
relationship with levels of trust in the market.
Demographics: In the West over the next decade, the large number or people retiring will drive
investment shifts, for example in the demand for retail rather than mutual funds, or for xed
income rather than equities.
Global economic cycles: The economic cycle appears to have been perturbed by the nature of the
current recession. The dynamics of growth, employment, savings, trade, and leverage may return
to previous cyclical behaviour or may instead follow a new pattern (prolonged recession, chaotic
behaviour). Linked to this, global imbalances may persist or resolve. These factors will affect the
demands placed on nancial markets in terms of volume, asset classes and acceptable levels of risk
and return. Global macroeconomic dynamics may also affect the process of globalisation and the
relative importance of nancial and ‘real’ markets.
Geopolitics: Growth rates over the next decade will powerfully inuence the structure of future
markets. A strong world economy will allow technological experimentation and new connections
among geopolitical regions and groupings. A faltering economy or, worse, one in a tailspin, would
be likely to lead to national retrenchment.
Technology
15
: This may lead to the creation of highly distributed trading platforms on which large
numbers of individuals carry out transactions. Individual mobile phone handsets, possibly receiving
live news and data feeds, may be used for trading; institutional trading strategies may also be
inuenced by communications on social networks. A new topology of either highly dispersed
exchanges or of interlinked international exchanges could take shape.
Loss/change of riskless (reference) assets: The current global asset ‘ecosystem’ uses the return to
riskless assets as a reference point for pricing risky assets. With sovereign debt now perceived to
carry risk, this point of reference may be on the verge of disappearing. The behaviour of nancial
markets without a commonly recognised riskless asset is uncertain, and it is not clear whether a
new common reference point will emerge. A connected issue is the link between sovereign debt
and national currencies, and the role of the dollar as the global reserve currency.
Asset classes: Products focusing on levels of risk exposure rather than dividends may become
more prominent; investment may shift from listed equity or derivatives towards synthetic
products and spread-betting. These may focus on the state-dependent pattern or returns rather
than ownership, and are likely to include more ‘exotic’ instruments. CBT may lead to the creation
of new nancial instruments for asset classes that are not currently directly traded using HFT or
algorithmic tools.
14 See also Figure 1.1. More detailed discussion of these drivers can be found in the workshop reports (Annex D refers).
15 In view of the importance of technology as a key driver for computer-based trading, Chapter 2 provides a more detailed
discussion. See also DR3 (Annex D refers).
23
The Future of Computer Trading in Financial Markets
Competition: Over and above technological changes, innovation in business models will shape
competitive dynamics. Features analogous to Amazon’s ‘other products you may like’ button may
be introduced into institutional trading products. Market shares and returns will be driven by
content, nancial products and costs. Firms may unbundle services to generate more commissions
or rebundle them to enhance investor lock-in. Exchanges are already increasing prots by
proposing value-added components; they could increasingly focus on content-driven models.
Change in (dis)intermediation: Technological and nancial market changes are altering both the size
and role of intermediaries. The pace, direction and implications of these shifts will depend on
whether such entities can operate across borders, the depth of funding that they inuence and
their impact on specic assets or investors. These developments are linked to CBT and HFT via
the arbitrage role of intermediaries. They may be implemented via CBT/HFT by the continuum
of CBT from algorithmic trading to HFT and by the degree to which the implications of CBT are
different for the price trading and asset management functions of intermediaries.
1.5 Structure of the Report
This Report is comprised of eight chapters. Chapter 1 provides the rationale for undertaking the
Project, sets out the aims and objectives, and its approach in terms of scope and content. In Chapter 2
technological developments are reviewed in detail, since this is a particularly important driver of change
affecting CBT. Their recent impact is reviewed and technology advances likely in the next ten years for
example, cloud computing, and custom silicon are explored. In Chapter 3 evidence on the impact of
CBT on key indicators of market quality including liquidity, price discovery/efciency and transaction
costs are assessed. The chapter begins by examining the evidence for past effects, and then considers
how impacts could change in the future, recognising that this will be contingent upon the mix of future
regulatory measures in place.
Chapter 4 examines the evidence for the impact of CBT on nancial stability, evidence for past effects
is reviewed and particular attention is given to the impact of self-reinforcing feedback loops in CBT
which can amplify internal risks and lead to undesirable interactions and outcomes in nancial markets.
The concept of ‘normalisation of deviance’, where unexpected and risky events come to be seen as
ever more normal, and its implications for nancial stability is also explored.
The issue of market abuse is examined in Chapter 5 from economic, regulatory and user perspectives.
It assesses the current impact of market abuse, and evidence on the perceptions of abuse using survey
data commissioned by the Project. Consideration is also given to how the relationship between market
abuse and CBT might evolve in the next ten years and possible courses of action to address the issue.
In Chapter 6 the potential economic impact of individual regulatory measures on stability, volatility
and also liquidity, price discovery/efciency and transaction costs are reviewed using a variety of new
studies commissioned by the Project. Benets as well as costs and risks are assessed. The measures
are diverse ranging inter alia from notication of algorithms, circuit breakers, minimum tick size
requirements and market maker obligations, to minimum resting times and periodic auctions.
The issue of how long term strategic factors, notably how CBT and HFT can affect trust and
condence in markets is discussed in Chapter 7. Particular emphasis is given to the role of rising
complexity, enabled by information technology, in nancial arrangements, transactions and processes
in recent decades, and also the supply of credit, in inuencing trust. It asks how can complexity be
constrained and reduced and highlights the potential for a greater role for standards in addressing
these issues.
Finally, Chapter 8 concludes the Report by drawing out the top level advice for policy makers, both
for the short term and long term. In the latter case, priorities for research and better data collection
are suggested.
24
2 The impact of technology
developments
Key findings
Continuing advances in the sophistication of ‘robot’
automated trading technology, and reductions in cost
are set to continue for the foreseeable future.
Today’s markets involve human traders interacting with
large numbers of robot trading systems, yet there is
very little scientic understanding of how such markets
can behave.
For time-critical aspects of automated trading, readily
customisable, special purpose silicon chips offer major
increases in speed; where time is less of an issue, remotely
accessed cloud computing services offer even greater
reductions in cost.
Future trading robots will be able to adapt and learn with
little human involvement in their design. Far fewer human
traders will be needed in the major nancial markets of
the future.
27
The Future of Computer Trading in Financial Markets
2 The impact of technology
developments
2.1 Introduction
The present-day move to ever higher degrees of automation on the trading oors of exchanges, banks
and fund management companies is similar to the major shift to automated production and assembly
that manufacturing engineering underwent in advanced economies during the 1980s and 1990s. This
trend is likely to have a corresponding effect on the distribution of employment in the nancial sector.
Already, a very large proportion of transactions in the markets are computer generated and yet the
characteristics and dynamics of markets populated by mixtures of human traders and machine traders
are poorly understood. Moreover, the markets sometimes behave in unpredictable and undesirable
ways. Few details are known of the connectivity network of interactions and dependencies in
technology enabled nancial markets. There is recognition that the current global nancial network
needs to be mapped in order to gain an understanding of the current situation. Such a mapping
exercise would enable the development of new tools and techniques for managing the nancial
network and exploring how it can be modied to reduce or prevent undesirable behaviour
1
. New
technology, new science and engineering tools and techniques will be required to help map, manage
and modify the market systems of the future.
2.2 How has financial market technology evolved?
The technology changes of the past ve years are best understood as a continuation of longer term
trends. Cliff et al. (DR3)
2
relate the history of technology in the nancial markets, briey covering the
18th, 19th and 20th centuries, and then explore in more detail the rapid and signicant changes which
have occurred at the start of the 21st century
3
.
The high speed processing of data and high speed communication of data from one location to another
have always been signicant priorities for the nancial markets. Long before the invention of computers
or pocket calculators, traders with fast mental arithmetic skills could outsmart their slower witted
competitors. In the 19th century, communication of nancially signicant information by messengers
on horseback was replaced by the faster ‘technology’ of carrier pigeons; then pigeons were made
redundant by telegraph; and then telegraph by telephones. In the last quarter of the 20th century,
the shift to computer-based trading (CBT) systems meant that automated trading systems could start
to perform functions previously carried out only by humans: computers could monitor the price of a
nancial instrument (for example, a share price) and issue orders to buy or sell the instrument if its
price rose above or below specied ‘trigger’ prices. Such very simple ‘program trading’ systems were
widely blamed for exacerbating the Black Monday crash of October 1987, the memory of which, for
several years afterwards, dampened enthusiasm for allowing computers to issue buy or sell orders in
the markets. Nevertheless, the real cost of computer power continued to halve roughly once every
two years (the so-called Moore’s Law effect), until by the late 1990s it was possible to buy, at no extra
cost in real terms, computers over 50 times more powerful than those used in 1987. All this extra
computer power could be put to use in implementing far more sophisticated processing for making
investment decisions and for issuing structured patterns of orders to the markets.
By the end of the 20th century, as the real cost of computing continued to fall at a dramatic pace,
the management of investment funds had become an increasingly technical eld, heavily dependent
1 The need to map, manage, and modify the nancial network is the central message of a speech Rethinking the Financial Network
by Haldane (2009).
2 DR3 (Annex D refers).
3 For other accounts of the recent history of technology developments in the nancial markets, the following three texts are
particularly recommended: Angel et al. (2010), Gomber et al. (2011) and Leinweber (2009). For a recent discussion of high
frequency trading, including interviews with leading practitioners, see Perez (2011) and Arnuk & Saluzzi (2012).
28
The impact of technology developments
on computationally intensive mathematical models to manage portfolio risk (i.e. to ‘hedge’ the risk in
the fund’s holdings). Manual methods for hedging risk were used for decades before the deployment
of electronic computers in fund management, but as the real cost of computing fell, and the speed
and capacity of computers increased, so computers were increasingly called upon to calculate results
that could guide the fund manager’s decisions to buy or sell, and to go ‘long’ or ‘short’. In this period,
growing numbers of funds based their investment decisions on so-called statistical arbitrage (commonly
abbreviated to ‘stat arb’). One popular class of stat arb strategies identies long-term statistical
relationships between different nancial instruments, and trade on the assumption that any deviations
from those long-term relationships are temporary aberrations; that the relationship will revert to its
mean in due course. One of the simplest of these ‘mean-reversion’ strategies is pairs trading, where
the statistical relationship which is used as a trading signal is the degree of correlation between just
two securities. Identifying productive pair-wise correlations in the sea of nancial market data is a
computationally demanding task, but as the price of computer power fell, it became possible to
attempt increasingly sophisticated stat arb strategies.
At much the same time, the availability of cheaper computation meant that it was possible to deploy
automated trading systems that were considered more ‘intelligent’ than those implicated in the 1987
crash. In most cases, this ‘intelligence’ was based on rigorous mathematical approaches that were
rmly grounded in statistical modelling and probability theory. The new wave of automated systems
concentrated on execution of a trade. The computer did not make the decision to buy or to sell
a particular block of shares or quantity of commodity, nor to convert a particular amount of one
currency into another: those decisions were still taken by humans (possibly on the basis of complex
statistical analysis). But, once the trading decision had been made, the execution of that trade was then
handed over to an automated execution system (AES). Initially, the motivation for passing trades to an
AES was that the human traders were then freed up for dealing with more complicated trades. As AES
became more commonplace, and more trusted, various trading institutions started to experiment with
more sophisticated approaches to automated execution: different methods and different algorithms
could be deployed to t the constraints of different classes of transaction, under differing market
circumstances; and hence the notion of ‘algorithmic trading’ (AT) was born
4
.
At the same time as AES systems were being developed to reduce market impact, other trading teams
were perfecting advanced stat arb techniques for identifying trading opportunities based on complex
statistical regularities which lay deep in the data: the price and volume data for hundreds or thousands
of instruments might have to be considered simultaneously and cross-compared in the search for
opportunities similar to the pairs trading of the 1980s, but typically involving many more than two
instruments. These advanced stat arb approaches were made possible by powerful computers, used to
run the statistical analyses, and also by developments in CBT infrastructure (the machinery which traders
use to communicate with each other and with the exchanges). Two notable developments were
straight-through processing (STP), where the entire trading process from initiation of an order to nal
payments and clearing is one seamless electronic ow of transaction processing steps with no human-
operated intermediate stages, and direct market access (DMA), where investors and investment funds
are given direct access to the electronic order books of an exchange, rather than having to interact
with the market via an intermediary, such as an investment bank or broker/dealer.
The convergence of cheap computer power, statistically sophisticated and computationally intensive
trading strategies and fast automated execution via STP and DMA, means that, in recent years, it
has become entirely commonplace for market participants to seek counterparties to a transaction
electronically, identify a counterparty and then execute the transaction, all within a few seconds. For
trades involving large quantities of a tradable instrument (so large that the volume of shares traded
One key aspect of modern automated trading systems is that they are designed to be highly autonomous: once switched on,
they are intended by design to ‘do their own thing’, at high speed, with little or no human intervention at the level of individual
trades. Although the computer running an algorithmic trading system can be switched out of action in an instant (for example,
by severing its power cable), in many real systems it is much more desirable to order the algorithmic trader to go into its
shutdown sequence, where it will do its best to sell off its portfolio of holdings as quickly as possible and with minimal losses
before actually switching itself off. This means that while humans still bear ultimate responsibility for writing and managing
algorithmic trading systems, humans are less and less involved in their minute-by-minute operation or control. An example of
the owners of an automated trading system being unable to re-establish effective control occurred on 1st August 2012, when
the US brokerage Knight Capital suffered massive losses as their automated trading system failed. See http://dealbook.nytimes.
com/2012/08/03/trading-program-ran-amok-with-no-off-switch/?nl=business&emc=edit_dlbkam_20120806 Accessed:
4 September 2012.
29
4
The Future of Computer Trading in Financial Markets
is likely to signicantly shift the price of the transaction – a phenomenon known as market impact),
the computerised search for a counterparty is not often conducted in the open, on a national stock
exchange, but instead takes place via the services of a private alternative electronic trading venue, a
‘dark pool’, which offers to match large volume counterparties in an anonymous or obfuscated fashion,
to try to reduce market impact. Details of the transaction are only reported to the national exchange
after it has been completed. Because many trading institutions see reducing market impact as a good
thing, the popularity of dark pools has increased signicantly in recent years.
The old ‘vertically integrated’ business model of investment banking is becoming increasingly
fragmented. One effect of the European Union’s (EU) Markets in Financial Instruments Directive (MiFID)
5
legislation was to create an ecosystem of small and medium-sized businesses offering ‘middleware’
technology components that could be purchased individually and then plugged together to achieve the
same functionality which had previously been the exclusive preserve of the trading systems developed
in-house by big institutions. This lowered the barriers to entry: with sufcient funding, one or two
entrepreneurs working in a rented ofce with a high speed internet connection can set up a trading
company and automate much, or perhaps all, of the workow required to run a fund.
At the same time, a new style of trading has emerged, known as high frequency trading (HFT)
6
,
where automated systems buy and sell on electronic exchange venues, sometimes holding a particular
position for only a few seconds or less. An HFT system might ‘go long’ by buying a quantity of shares
(or some other nancial instrument, such as a commodity or a currency) hold it for perhaps two or
three seconds, and then sell it on to a buyer. If the price of the instrument rises in those two or three
seconds, and so long as the transaction costs are small enough, then the HFT system has made a prot
on the sale. The prot from holding a long position for three seconds is unlikely to be great, and may
only amount to a few pennies but if the HFT system is entirely automated, then it is a machine that
can create a steady stream of prot every hour that the market is open. A recent study
7
indicated that
the total amount of money extractable from the markets via HFT may be more modest than might be
supposed: a few tens of billions of dollars in the US markets. Despite this, the low variation in positive
returns from a well-tuned HFT system is an attractive feature and one that makes HFT an area of
intense interest in the current markets. Of course, even if the total prots extractable via HFT really
are limited, the downside risks, the total potential worst case losses and costs that might be incurred if
HFT technology fails, could in principle be much larger. There is a concern that some CBT systems, like
other novel trading systems in the past, could be making a steady stream of small prots but at the risk
of causing very big losses if (or when) things go wrong. Even if each individual CBT system is considered
to be stable, it is well known that groups of stable systems can, in principle, interact in highly unstable
ways: the systemic risks of CBT are currently not well understood and are discussed in more detail in
Chapter 4.
As the global nancial markets became dependent on computers running automated trading systems
and communicating with each other over optical bre networks
8
, the speeds of computation and of
communication became two of the primary means by which competitive advantage could be gained
and maintained. The effect of this in present day markets is discussed in the next section.
2.3 What are the key current technology developments?
2.3.1 From past to present
Firms at the front line of the nancial markets, such as investment banks, fund management
companies and exchange operators, are all critically dependent on information technology and the
telecommunications networks that allow computers to communicate with each other. For the past
5 MiFID is an EU law that came into force in April 2004. In 2011, EU legislators began circulating drafts and revisions of a second
version of MiFID, known as MiFID II.
6 Google Trends indicates that Google’s users, worldwide, have only used the phrase ‘high frequency trading’ as a search term in
the past four years, and ‘algorithmic trading’ only in the past six years. See www.google.com/trends Accessed: 17 September 2012.
7 Kearns et al. (2010).
8 The transmission speed of bre-optic communications (i.e. via pulses of light beamed along glass bres) is not quite equal to
c, the speed of light in a vacuum. Transmission by line-of-sight microwave links through the atmosphere is slightly faster than
bre-optic communications and establishing microwave communications networks for HFT is currently an area of intense
business activity. See Troianoivski (2012).
30
The impact of technology developments
two decades, nearly all such rms used their own in-house information technology systems, very often
involving powerful server computers connected to ‘client’ computers running on the desks of each
employee. Almost always, the client computers would be standard personal computers (PC), of the
sort available from high-street retailers, and the server computers would actually be constructed from
several very high specication PCs, all located together and connected to each other in a single room;
that room being the rm’s ‘server room’ or ‘data centre’.
As Cliff et al. (DR3) describe in some detail, the global information technology industry is currently
undergoing a major shift towards cloud computing, where ultra large scale data centres (vast
warehouses full of interconnected computers) are accessed remotely as a service via the internet,
with the user of the remotely accessed computers paying rental costs by the minute or by the hour
(See Figure 2.1). This greatly reduces the cost of high performance computing (HPC), and hence lowers
barriers to entry for individuals or rms looking to use supercomputer scale HPC for the automated
design and optimisation of trading systems. Rather than spending millions of dollars of capital
expenditure on an in-house HPC data centre facility, it is now possible to obtain the same results by
renting HPC from cloud computing providers for a few thousand dollars. It is no longer necessary to
have the nancial resources of a major hedge fund or investment bank to engage in development of
highly technology-dependent approaches to trading. The full implications of this are not yet clear.
Figure 2.1: Inside a cloud computing data centre, large numbers of individual computers
are stacked in racks and any number of them can be remotely accessed via the internet as
and when users require them. Typically, users pay only a small rental cost for each hour’s
access to each computer, so super-computer levels of processing power can be accessed at
comparatively low cost.
At the same time, the desire for ultra high speed processing of nancial data has led a number of
market leaders to abandon the use of general purpose computers, such as commercially available
PCs, and replace them instead with customised special purpose silicon chips. Some of these silicon
chips have hundreds or thousands of independent small computers on them, each operating in
parallel, offering huge increases in speed. This technology and the companies which have adopted it
are discussed in more depth in DR3, which concludes that the switch to such ‘custom silicon’ is set to
continue in the coming decade.
31
The Future of Computer Trading in Financial Markets
These two trends mean that greater computational power and greater speed are becoming more
readily available per unit cost. Technologists have therefore turned their attention to developing
innovative new systems for automated generation of trading decisions and/or automated execution of
the orders necessary to enact those decisions.
One major new technology that is currently the focus of signicant research and development is the
prospect of computers being programmed to ‘understand’ not only the numeric information of market
prices, volumes and times, but also the non-numeric semantic information that is carried in human-
readable data streams (such as written news reports, status updates and ‘tweets’ on social media
websites) and audio data (such as telephone calls, radio shows, podcasts and video sequences). This
is an issue explored in depth by Mitra et al. (DR8), whose work is discussed in Section 2.3.2
9
.
Despite the increases in computer power, processing speed and sophistication of computer algorithms,
the present day nancial markets still involve large numbers of human traders. There are good reasons
to expect that, for the next decade or so, the number of human participants in the market will remain
signicant. For most major markets in the USA, UK and mainland Europe, the proportion of computer-
generated trades is estimated to be variously 30%, 60% or more
10
. What is clear is that the current
markets involve large numbers of human traders interacting with large numbers of automated trading
systems. This represents a major shift in the make-up of the markets and may well have affected their
dynamics (see Chapters 1 and 7). Research that explores the interactions of humans and algorithmic
trading systems is discussed in Section 2.3.3.
2.3.2 Automated analysis of market news and sentiment
Mitra et al. (DR8) consider which asset classes are best suited to automated trading by computers.
They contend that different nancial instruments have different liquidity and that the optimal trading
frequency for an instrument can be expressed as a function of the liquidity of its market, amongst
other factors. The higher the optimal trading frequency, the more useful AT is. The traders in a market
can be classied as one of two types: those who aim to prot by providing liquidity (so-called inventory
traders) and those who aim to prot by trading on the basis of information (so-called informed or
value motivated traders). Inventory traders act as market makers: they hold a sufciently large quantity
of an instrument (their inventory) that they are always able to service buy or sell requests, and they
make money by setting a higher price for selling than for buying (this is the type of business model that
is familiar in any airport currency exchange retail outlet). Inventory traders can, in principle, operate
protably without recourse to any information external to the market in which their instruments are
being traded.
The informed/value motivated traders, make use of information in news stories and related discussion
and analysis to form a view about what price an instrument should be trading at, either now or in
the future, and then buy or sell that instrument if their personal opinion on the price differs from the
current market value. In recent years, technologies have been developed that allow computers to
analyse news stories and discussions on social networking websites, and these technologies are rapidly
increasing in sophistication.
In DR8, Mitra et al. argue that the primary asset classes suitable for automated trading are equities,
including exchange traded funds (ETFs) and index futures, foreign exchange (FX) and, to a lesser
extent, commodities and xed income instruments. ETFs are securities traded on major exchanges as
if they were standard stock equities (shares in a company), but the ETF instead represents a share in a
holding of assets, such as commodities, currency or stock. As would be expected, news events (both
anticipated and unanticipated) can affect both traditional manual trading for these asset classes and also
automated trading activities. Anticipated events are those such as the release of ofcial ination data
by government treasury departments or scheduled earnings announcements by rms; unanticipated
events are those such as news concerning major accidents, terrorist actions or natural disasters.
Because of the effect that news events can have on the prices of nancial instruments, major global
9 DR8 (Annex D refers).
10 Accurate estimates of the volume of high frequency trading are difcult to obtain, and in any case are contingent on the
precise denition used. For estimates, see Kaminska (2011), Kaminska (2009) and http://www.tradersmagazine.com/news/high-
frequency-trading-benets-105365-1.html?zkPrintable=true Accessed: 3 September 2012.
32
The impact of technology developments
companies exist to provide news feeds specic to the nancial markets, including Thompson Reuters,
the Financial Times, the Wall Street Journal, Dow Jones Newswire and Bloomberg. Much of the news
content comes in formats that can readily be processed by computers. The content from traditional
mass-market news broadcasters, such as the BBC, can also be processed by computers (possibly after
some automated reformatting or conversion from audio/video to text-based transcripts).
Because of these developments, researchers in academia and in nancial institutions have developed
methods for news analytics. Signicant advances have been made in recent years and the techniques
are still growing in sophistication. In general, it is reasonable to predict that a computer will be able to
react to a breaking news story faster than a human can but, of course, this is only useful if its analysis
of the story is actually correct. Some practitioners argue that automated trading and news analytics
puts manual (human-based) trading at a considerable disadvantage and that this applies to both retail
investors and institutional investors. Although news analytics technology cannot yet reliably outperform
a well-informed human trader reading the same material, and has only very limited abilities in
comparison to the human capacity for reasoning and lateral thinking, the capabilities and sophistication
of news analytics systems will continue to increase over the next decade, possibly to the point where
they surpass the performance of human analysts and traders.
2.3.3 Interactions between human and algorithmic trading systems
The global nancial markets are now populated by two types of economic agents: human traders and
‘software agents’. The latter are either algorithmic systems performing trading jobs that 10 or 20 years
ago would have been the responsibility of humans, or HFT systems doing jobs that no human could
ever hope to attempt. Interactions between human traders in electronic markets have long been
studied in the eld known as experimental economics, and more recently the interactions between
software agent traders in electronic markets have been the topic of various research studies in so-
called agent-based computational economics. Curiously, these two research elds are largely distinct:
the rst studies markets populated entirely by human traders while the second studies markets
populated entirely by algorithmic software agent traders. There is surprisingly little scientic literature
that explores heterogeneous markets, populated by both humans and algorithmic systems. In DR13
11
,
De Luca et al. provide a survey of the peer reviewed published research.
The rst study to report any results was published in 2001
12
by a team of researchers working at IBM,
where two trading algorithms were each demonstrated to outperform human traders. IBM’s work
served as an inspiration to Professor Jens Grossklags of the University of California at Berkeley, who
used similar methods to explore different questions in studies published in 2003 and 2006
13
. Until 2011,
the IBM experiments and Grossklags’ work were the only three peer reviewed papers in the scientic
literature that studied this topic. This is a startling gap in the research literature; one that has only very
recently started to be lled.
In DR13, there is a detailed description and analysis of results from several preliminary experiments,
conducted specically for the Foresight Project. In DR25
14
, this experimental work is replicated and
extended. In the latter, some of the articial experimental constraints that were used in earlier work
are relaxed, for greater realism and hence increased relevance to the real world markets. The key
conclusion from these preliminary experiments in DR13 and DR25 is that the previously reported
outperformance of algorithmic trading systems in comparison to human traders may be due simply to
the fact that computers can act and react faster than humans. This provides empirical support for the
intuitive notion that the primary advantage that current software agent trading algorithms have over
humans is the speed at which they operate, although being faster at trading does not necessarily lead to
greater overall efciency in the market.
11 DR13 (Annex D refers).
12 Das et al. (2001).
13 Grossklags & Schmidt (2006).
14 DR25 (Annex D refers).
33
The Future of Computer Trading in Financial Markets
Johnson et al. (DR27)
15
argue that analysis of millisecond-by-millisecond US stock price movements
between 2006 and 2011 suggests the existence of a step change or phase transition in the dynamics
and behaviour of nancial markets, in which human traders and automated algorithmic ‘robot’ trading
systems interact freely. Above a particular time threshold, humans and algorithmic systems trade with
one another; below the threshold, there is a sudden change to a market in which humans cannot
participate and where all transactions are robot-to-robot. In addition to this, results in DR25 from a
series of human-versus-robot experimental nancial markets broadly support the hypothesis in DR27
concerning this change in market dynamics.
2.3.4 From the present to the future
Some key issues in today’s markets look likely to remain vitally important in the future. For
instance, cyber security will remain a core concern. Electronic attacks on the computer systems
and communications networks of the global nancial markets are always likely to be attractive to
criminals. Furthermore, the widespread move towards greater reliance on advanced computing
technology means that the speed of light is now a signicant limiting factor in determining how trading
systems interact with one another. Even with the best technology imaginable, information cannot be
transmitted faster than the speed of light, and even at light speed it does actually take measurable
periods of time to move information across an ocean, or even across a few city blocks. On the
assumption that Einstein was right, this constraint is immutable.
The evolution of classes of algorithms is discussed in detail by Gyurkó in DR18
16
. Gyurkó concludes that
the hunt for trading prots induces extreme competition, with any one rm’s competitive advantage
from a technology innovation being quickly eroded over time as competitors innovate and catch up,
and hence market participants are forced to invest constantly in research and technology development.
Recent academic research indicates that the next generations of trading algorithms will be adaptive
(learning from their own experience) and will be the result of automated computerised design and
optimisation processes. Therefore, the performance of these algorithms may be extremely difcult to
understand or explain, both at the level of an individual algorithm and at the system level of the market
itself. Admittedly, knowledge about the functioning of nancial markets populated entirely by humans is
imperfect, even after hundreds of years; but present day markets with high levels of automation may be
sufciently different from human-only markets that new lessons need to be learned.
2.4 Technology advances likely in the next ten years
As with many such competitive interactions, in the technology ‘arms race’ new innovations only confer
competitive advantage to an innovator for as long as it takes the innovator’s competitors to copy that
innovation or to come up with counter-innovations of their own. As soon as all traders are using a
particular new technology, the playing eld is levelled again
17
. Nevertheless, several of the present day
technology trends seem likely to remain signicant factors over the next decade.
2.4.1 Cloud computing
Cloud computing offers the possibility that it is no longer necessary to have the nancial resources of
a major hedge fund or investment bank to engage in development of highly technology-dependent
approaches to trading. Nevertheless, there are regulatory and legislative issues that need to be carefully
examined. For example, for jurisdictional reasons the geographic location of the remote servers can
matter greatly. Cloud computing service providers are well aware of such concerns and can offer
geographic guarantees in their service level agreements and contracts. Moreover, remote access
of computing facilities, even at the speed of light, means that there will be latencies in accessing the
remote systems. For many applications these may not matter, but for trading activities the latencies
inherent in communicating with remote data centres can be prohibitive. Latency would certainly be a
15 DR27 (Annex D refers).
16 DR18 (Annex D refers).
17 The arms race nature of adaptive trading technology development, and the steadily reducing levels of human involvement in
major markets, means that algorithmic game theory may be a valuable tool in understanding systemic issues in automated
nancial markets. The well-known Prisoners’ Dilemma game theory problem demonstrates that the stable outcome of rational
interactions among competing self-interested entities is not necessarily the one that maximises the entities’ rewards, i.e. the
outcome may not be welfare enhancing.
34
The impact of technology developments
problem if an institution tried to run its automated HFT algorithms ‘in the cloud’, but it is important
to remember that not all trading is HFT. There are other modes of trading, such as long-only macro
trading
18
, that are not so latency sensitive.
The primary impact of cloud computing on activities in the nancial markets in the next ten years
will not be in the provision of computing facilities that automate execution, but rather in the ability
of the cloud to provide cheap, elastically scalable HPC. Such cheap, remote HPC will allow massively
computer-intensive procedures to be deployed for the automated design and optimisation of trading
strategies and execution algorithms: this computational process is not latency sensitive. Many major
investment banks and hedge funds already own and operate private data centres, but they do this
for business critical operations and only a fraction of the capacity can be assigned to HPC uses. The
ability to either extend existing in-house computing power by adding on externally provisioned cloud
based resources, or to simply outsource all of the HPC provisioning to a cloud provider, opens up new
possibilities that are only just being explored and lowers barriers to entry.
2.4.2 Custom silicon
General purpose, commercially available PCs have, over recent years, moved from being based on a
single central processing unit (CPU) chip to a new breed of CPU chips that have multiple independent
computers (known as ‘cores’) built into them. Perusal of high-street stores will reveal PCs with dual-
core and with quad-core chips as standard. In rough terms, a dual-core chip can do twice as much
work as a single-core chip per unit of time, and a quad-core can do four times as much. Currently,
there is a major shift underway towards so-called many-core computing, exploring the increased speed
offered by using chips with many tens or hundreds of independent cores operating in parallel. This
often involves using specialised processing chips originally designed for computer graphics processing.
Furthermore, as noted above, the desire for ultra high speed processing of nancial data has led to a
move within the industry to replace general purpose computers with special-purpose silicon chips that
can be customised or programmed ‘in the eld’ (i.e. the end-user of the chip customises it to t that
user’s needs).
Currently, the most popular technology of this type is a chip known as a eld-programmable gate
array (FPGA). Currently, programming an FPGA is a very complicated and time consuming task: the
programmer has to translate an algorithm into the design for an electronic circuit and describe that
design in a specialised hardware description language. Despite these complexities, the switch to such
custom silicon is likely to continue over the next decade because of the speed gains that it offers
19
.
Over that period it is probable that the use of FPGAs will be supplanted by a newer approach to
custom silicon production, involving more readily eld-programmable multi-core or many-core chips.
Such chips will be programmable in a high level software language, much like current industry standard
programming languages. This means that conventionally trained programmers can write algorithms
that are then ‘compiled down’ onto the underlying silicon chip hardware, without the need to learn
specialised FPGA hardware description languages. This has the potential to reduce custom silicon
development times (currently measured in days or weeks) to only a few minutes from describing
a trading algorithm in a high level programming language, to having that algorithm running on a
parallel high speed computing array composed of a multitude of independent customised silicon chip
processing elements. In DR3, it is suggested that this style of computer hardware is likely to be in wide
use within ten years. This type of hardware would have enough computing power to enable future
generations of trading algorithms that are adaptive (learning from their own experience) and that will
not have been designed by human engineers, but rather will be the result of automated computerised
design and optimisation processes.
18 Long-only macro trading is where a trader maintains a portfolio of holdings that are bought only in the expectation that their
market value will increase (that is, the trader is taking long positions, as opposed to short positions where the trader would
benet if the market value decreases); and where the trader’s alterations to the portfolio are driven by macroeconomic
factors, such as national interest rates, which tend to alter relatively slowly, rather than the second-by-second uctuations
commonly exploited by high frequency traders.
19 A little more than a year after review DR3 was written, in May 2012, the trading technology provider company Fixnetix
announced that they had delivered new products based on FPGA technology to a major bank in New York. The bank wishes
to remain anonymous. Fixnetix stated that the FPGA-based hardware executed pre-trade compliance and 20 customisable
risk checks “in nanoseconds”. That is, the new hardware took less than one microsecond (less than one millionth of a second)
to execute these pre-trade checks. See http://bit.ly/LN4seW Accessed: 17 September 2012.
35
The Future of Computer Trading in Financial Markets
2.4.3 Computer designed trading algorithms that adapt and learn
As De Luca et al. discuss at length in DR13, researchers have been studying and rening adaptive
trading algorithms since the mid-1990s and, in 2001, IBM showed two such algorithms to be capable
of outperforming human traders. Adaptive trading algorithms are automated trading systems that can
learn from their experience of interacting with other traders in a market, improving their actions over
time, and responding to changes in the market. Since the late 1990s, researchers have also studied
the use of automated optimisation methods to design and improve adaptive trading algorithms. In
automated optimisation, a vast space of possible designs is automatically explored by a computer
program: different designs are evaluated and the best-performing design found by the computerised
search process is the nal output. Thus, new trading algorithms can be designed without the
involvement of a human designer
20
. The use of these techniques in the nance industry looks likely to
grow over the next decade. This is a development that is enabled and accelerated by the step change
drop in cost of HPC offered by cloud computing service providers, and by the huge speed increases
offered by custom silicon.
Because these next generation trading algorithms may have had little or no human involvement in their
design and renement, and because they operate at truly superhuman speeds, the behaviour of any
one such automated trader may be extremely difcult to understand or explain, and the dynamics of
markets populated by such traders could be extremely difcult to predict or control. The concept of
adaptive trading algorithms, whose behaviour may be difcult to understand, and whose dynamics in
conjunction with other traders, whether automated or not, may be difcult to predict or control, will
be quite disturbing. One natural instinct is to try to ban all such algorithms from one’s own market.
But that will not work, partly because the denition of what constitutes an adaptive algorithm is itself
ill dened, and partly because markets will either exist, or can easily be set up, which will allow such
trading practices. Once such markets exist, and they will, any market crashes or other disturbances in
prices will reverberate directly back to the markets in which such algorithms are banned. Rather the
need is to develop a culture and understanding which serves to limit such dangers.
Other studies, such as those discussed in DR25 and DR27, are needed to increase our understanding
of market dynamics in the presence of automated trading systems that are adapting over time
and whose designs are the result of computerised optimisation processes. Furthermore, as Cliff &
Northrop describe at length in DR4
21
, there are likely to be valuable lessons to be learned from
elds other than nance or economics. In particular, sociological studies of how people interact with
technology that is ‘risky’, in the sense that its safe operating limits are not fully understood in advance,
have revealed that it is important to be mindful of the dangers of falling into a pattern of behaviour,
a process called ‘normalisation of deviance’. When a group of people engage in this process, they
gradually come to see events or situations that they had originally thought to be dangerous or likely
to cause accidents as less and less deviant or dangerous, and more and more normal or routine. This
greatly increases the risk of major accidents. Normalisation of deviance is discussed in more depth in
Chapter 4.
Normalisation of deviance can be avoided by adopting practices from what are known as high reliability
organisations (HROs). Examples of HROs that have been studied include surgical teams, teams of
reghters and aircraft-carrier ight deck operations crews. In all such groups, safety is a critical
concern, and it has been found that HROs have often independently developed common deviance
monitoring processes involving careful post-mortem analyses of all procedures, even those in which
no problems were apparent, conducted condentially but with an internal atmosphere of openness
and absence of blame. Adoption of HRO working practices and other approaches developed in safety-
critical engineering (such as the use of predictive computer simulation models, as discussed in more
depth in DR4, DR14, and DR17
22
) would help to mitigate possible ill effects of increased reliance on
risky technology in the nancial markets (for further discussion see Chapter 4).
36
20 Cliff (2009).
21 DR4 (Annex D refers).
22 DR14 and DR17 (Annex D refers).
The impact of technology developments
2.5 Conclusions
It is reasonable to speculate that the number of human traders involved in the nancial markets could
fall dramatically over the next ten years. While unlikely, it is not impossible that human traders will
simply no longer be required at all in some market roles. The simple fact is that we humans are made
from hardware that is just too ‘bandwidth-limited’, and too slow, to compete with new developments
in computer technology.
Just as real physical robots revolutionised manufacturing engineering, most notably in automobile
production, in the latter years of the 20th century, so the early years of the 21st seem likely to be
a period in which a similar revolution (involving software robot traders) occurs in the global nancial
markets. The decline in the number of front line traders employed by major nancial institutions in
the past few years as automated systems have been introduced is likely to continue over the next
few years.
Nevertheless, there may be increased demand for developers of automated trading systems, and
for designers of customised computer hardware that runs at high speed and with low latencies. It is
most likely that the skills needed in these designers and developers will be those learned in advanced
undergraduate and postgraduate degree courses. From a UK perspective, serious investment in
coordinated programmes of research and development (for example, funded by the UK research
councils or the UK Technology Strategy Board)
23
could help to secure the future ‘talent base’ (i.e.
the pool of trained scientists and engineers that have skills appropriate for work on advanced trading
algorithms and hardware).
The increased reliance on CBT is certainly not without its risks. Sudden market crashes (or sudden
bubbles) in the prices of nancial instruments can occur at greater speed, with chains of events
proceeding at a much faster pace than humans are naturally suited to deal with. Furthermore, the
globally interconnected network of market computer systems arguably means that an adverse event in
one market now has greater potential to trigger a wave of contagion that could affect markets around
the world (see Chapter 4 for further discussion of systemic risks of CBT). Equally, natural hazards in
the form of oods or electrical storms can incapacitate data centre telecommunications networks; an
interplanetary coronal mass ejection (a huge burst of plasma and electromagnetic energy from the sun,
causing a geomagnetic storm when the ejection reaches the Earth) could seriously disrupt or disable
the electrical power systems and radio communications of an entire city or larger area. Ensuring the
resilience of critical computer and communications systems in the face of such major natural hazards,
and in the face of attack from cyber criminals, must always remain a priority.
Even in the absence of such exogenous hazards, there are serious issues to be addressed in dealing
with major endogenous risks, such as destabilising systemic internal feedback loops, and the pernicious
effect of normalisation of deviance in risky technology systems, both of which are discussed further
in Chapter 4. In addressing issues of systemic risk and nancial stability, it is particularly important for
regulators to develop and maintain the capacity to analyse extremely large data-sets. The challenge of
dealing with ‘big data’ is already severe enough for any single major nancial trading venue (for example,
an exchange or dark pool): the number of active HFT systems and the rate at which they generate (and
cancel) orders means that the venue’s order book showing the best bid and offer for a particular stock
may need to be updated thousands of times per second. A slight shift in the price of that stock can
then cause the prices of many tens or hundreds of derivative contracts, such as options or exchange
traded funds, to need to be recalculated. For regulators, the issue is made even more difcult as they
are required to deal with aggregate market data generated simultaneously by multiple trading venues,
as part of their regulatory role to oversee fair and orderly market systems, to identify policy violations
and to monitor the need for policy revisions. Well designed mechanisms to manage the risk of volatility
are one means of reducing the effects of risk in today’s and future markets.
23 The UK does already have signicant investment in relevant PhD-level research and training, with doctoral students from
various initiatives such as the UK Doctoral Training Centre in Financial Computing and the UK Large-Scale Complex IT
Systems Initiative (www.nancialcomputing.org and www.lscits.org, respectively) working on issues of direct relevance to the
topics addressed in this chapter. Despite this, there has been comparatively little investment in coordinated programmes of
relevant academic research among teams of post-doctoral level research workers; that is, major research projects, requiring
more effort than a single PhD student working for three years do not seem to have been prioritised.
37
The Future of Computer Trading in Financial Markets
On the basis of the evidence reviewed in this Chapter, it is clear that both the pace of development
of technology innovations in nancial markets and the speed of their adoption look set to continue
or increase in the future. One important implication of these developments is that trading systems
can today exist anywhere. Emerging economies, such as those of Brazil, Russia, India and China, may
capitalise on the opportunities offered by the new technologies and in doing so may possibly, within
only a few decades, come to rival the historical dominance of major European and US cities as global
hubs for nancial trading.
The next chapter builds on the information provided here to assess the impact of CBT on liquidity,
price efciency/discovery and transaction costs.
38
3 The impact of computer-based
trading on liquidity, price efficiency/
discovery and transaction costs
Key findings
The past ten years have been a difcult period for investors
in European and US nancial markets due to low growth
and a succession of crises. Nevertheless, computer-based
trading has improved liquidity, contributed to falling
transaction costs, including specically those of institutional
investors, and has not harmed market efciency in regular
market conditions.
The nature of market making has changed, shifting from
designated providers to opportunistic traders. High
frequency traders now provide the bulk of liquidity, but
their use of limited capital combined with ultra-fast speed
creates the potential for periodic illiquidity.
Computer-driven portfolio rebalancing and deterministic
algorithms create predictability in order ows. This allows
greater market efciency, but also new forms of market
manipulation.
Technological advances in extracting news will generate
more demand for high frequency trading, leading to
increased participation in high frequency trading, limiting
its protability.
However, there are some issues with respect to periodic
illiquidity, new forms of manipulation and potential threats
to market stability due to errant algorithms or excessive
message trafc that need to be addressed.
41
The Future of Computer Trading in Financial Markets
3 The impact of computer-based
trading on liquidity, price efficiency/
discovery and transaction costs
3.1 Introduction
Technology has transformed asset markets, affecting the trading process from the point of asset
selection all the way through to the clearing and processing of trades. Portfolio managers now use
computerised order management systems to track positions and determine their desired trades, and
then turn to computerised execution management systems to send their orders to venues far and
wide. Computer algorithms, programmed to meet particular trading requirements, organise orders to
trade both temporally across the trading day and spatially across markets. High frequency traders use
ultra-fast computing systems and market linkages both to make and take liquidity across and between
markets. Transaction cost analysis, using computers to capture price movements in and across markets,
then allows asset managers to calculate their trade-specic transaction costs for particular trading
strategies, and to predict their costs from using alternative strategies.
What is particularly striking is that virtually all of these innovations have occurred within the past ten
years
1
. In this short interval, the market ecology has changed, with markets evolving from traditional
exchange-based monopolies to networks of computer-linked venues
2
. Yet, while the process of trading
in markets has changed, the function of markets remains the same: markets provide liquidity and
price discovery that facilitates the allocation of capital and the management of risk. The purpose of
this chapter is to assess the evidence on the impact of computer trading on liquidity, transaction costs
and market efciency. The likely future evolution of computer trading on these dimensions of market
quality is also considered. The implications of these recent and potential developments for policy
makers and regulators are discussed in Chapter 8.
3.2 Determining the impact of computer-based trading on liquidity, price
efficiency/discovery and transaction costs
Determining the impact of technology on market quality (the general term used to describe the
liquidity, transaction costs and price efciency of a market) is complicated by the many ways in which
computers affect the trading process. Moreover, there is not even complete agreement on how to
dene some of these technological innovations. High frequency trading (HFT) is a case in point. HFT
was virtually unknown ve years ago, yet it is estimated that high frequency traders in the USA at times
participate in 60% or more of trades in equities and futures markets
3
.
The US Securities and Exchange Commission (SEC) 2010 Concept Release on Equity Market Structure
(SEC (2010)) describes HFT as employing technology and algorithms to capitalise on very short-lived
information gleaned from publicly available data using sophisticated statistical, machine learning and
other quantitative techniques. Yet, even within this general description, the SEC notes the difculty in
characterising what HFT actually means:
The term is relatively new and is not yet clearly dened. It typically is used to refer to
professional traders acting in a proprietary capacity that engage in strategies that generate
a large number of trades on a daily basis (…) Other characteristics often attributed
to proprietary rms engaged in HFT are: (1) the use of extraordinarily high speed and
42
1 See DR5 for discussion of the origins and growth of computer-based trading (Annex D refers).
2 See DR6 for discussion (Annex D refers).
3 Accurate estimates of the volume of high frequency trading are difcult to obtain, and in any case are contingent on the
precise denition used. For estimates, see Kaminska (2011), Kaminska (2009) and http://www.tradersmagazine.com/news/high-
frequency-trading-benets-105365-1.html?zkPrintable=true Accessed: 3 September 2012.
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
sophisticated computer programs for generating, routing, and executing orders; (2) use
of co-location services and individual data feeds offered by exchanges and others to
minimize network and other types of latencies; (3) very short time-frames for establishing
and liquidating positions; (4) the submission of numerous orders that are cancelled shortly
after submission; and (5) ending the trading day in as close to a at position as possible
(that is, not carrying signicant, unhedged positions over night)
4
.
Despite the lack of clarity as to the exact meaning of HFT, there is little disagreement regarding its
importance in markets. Many high frequency traders act as market makers by placing passive limit
orders onto electronic order books
5
. These passive orders provide the counterparty for traders
wishing to nd a buyer or seller in the market. In addition, high frequency traders often engage in
statistical arbitrage, using their knowledge of correlations between and within markets to buy an
asset trading at a low price and simultaneously sell a correlated asset trading at a higher price. This
activity essentially ‘moves’ liquidity between markets, providing a new dimension to the market making
function. The centrality of this role means that high frequency traders are involved in a large percentage
of market volume. Estimates of HFT involvement in US equity trading can be as high as 60%, with
estimates of HFT involvement in European equities markets ranging from 30–50%
6
. Estimates of HFT
in futures foreign exchange (FX) markets are in a similar range. Prots of high frequency traders in
the USA in 2009 have been estimated at $7.2 billion
7
, although others argue that the actual gure is
much lower
8
. By comparison, the total value of electronic order book share trading on the NYSE and
NASDAQ during 2009 was in excess of $30 trillion
9
.
There are some important differences between such high frequency market making and its traditional
specialist-based counterpart. HFT market makers rely on high speed computer linkages (often
achieved by co-locating their computers at the exchange) to enter massive numbers of trades with
the goal of earning the bid-ask spread. Such traders generally hold positions for very short periods
of time (in some cases, microseconds) and some operate with very low levels of capital, whereas
specialists in traditional markets have obligations to stand ready to buy and sell. HFT market makers
trade opportunistically: they typically do not hold large inventory positions and they manage their risks
by curtailing trading when market conditions are too adverse
10
. This behaviour raises the spectre of
periodic illiquidity.
The debate surrounding HFT has become increasingly heated, reecting the varied perspectives on the
ability (and desirability) of high frequency traders to move faster (and on the basis of potentially greater
information) than other traders
11
. Paul Krugman represents the contrarian view of HFT:
It’s hard to imagine a better illustration [of social uselessness] than high frequency trading.
The stock market is supposed to allocate capital to its most productive uses, for example
by helping companies with good ideas raise money. But it’s hard to see how traders
who place their orders one-thirtieth of a second faster than anyone else do anything to
4 See US Securities and Exchange Commission (2010).
5 See DR10 and DR12 (Annex D refers).
6 Accurate estimates of the volume of high frequency trading are difcult to obtain, and in any case are contingent on the
precise denition used. For estimates, see Kaminska (2011), Kaminska (2009) and http://www.tradersmagazine.com/news/high-
frequency-trading-benets-105365-1.html?zkPrintable=true Accessed: 3 September 2012.
7 See http://online.wsj.com/article/SB10001424053111904253204576510371408072058.html Accessed: 6 September 2012.
8 Kearns et al. (2010).
9 This is the total number of shares traded through the exchange’s electronic order book multiplied by their respective matching
prices. See World Federation of Exchanges: http://www.world-exchanges.org/statistics/annual Accessed: 17 September 2012.
10 It should be noted that traditional specialists (before the advent of CBT) also typically avoided holding large inventory
positions. The number of specialists obliged to make markets has dwindled in the face of new competition from CBT. On the
London Stock Exchange (LSE) there are ofcial market makers for many securities (but not for shares in the largest and most
heavily traded companies, which instead trade on an automated system called TradElect). Some of the LSE’s member rms
take on the obligation of always making a two-way price in each of the stocks in which they make markets. Their prices are the
ones displayed on the Stock Exchange Automated Quotation (SEAQ) system and it is they who generally deal with brokers
buying or selling stock on behalf of clients.
11 High frequency traders generally use proprietary data feeds to get information on the state of the market as quickly as
possible. In the USA, this means that they receive information before it is delivered over the consolidated tape, raising issues
of fairness and potential harmful effects on the cost of capital. See US Securities and Exchange Commission (2010).
43
The Future of Computer Trading in Financial Markets
improve that social function … we’ve become a society in which the big bucks go to bad
actors, a society that lavishly rewards those that make us poorer
12
.
Rhetoric aside, while the issues surrounding computer-based trading (CBT), and HFT in particular, are
complex, they are amenable to economic analysis. A useful starting point is to consider how market
quality has fared as this new market ecology has developed
13
.
3.3 Past: what has been the impact of computer-based trading on market
quality in recent years?
In this section the evidence on the effects on market quality of CBT is surveyed with emphasis being on
the highest quality peer reviewed work. However, CBT is a relatively recent phenomenon and there is
not much published research available. The eld is evolving and so some caution is required in drawing
conclusions. By way of comparison, the issue of whether smoking has negative health outcomes has
concluded with almost universal agreement based on extensive, careful research. As the research on
CBT is so recent, it has not yet achieved that level of agreement. Some discussion of methodological
issues is presented in Box 3.1.
Box 3.1: Methodology
Some of the methodological issues relating to measurement and statistical inference considered in
this chapter are discussed briey here.
Measurement of HFT
There are two approaches to the measurement of CBT or, more specically, HFT:
1. To use proxies (like message trafc) that are related to the intensity of trading activity. An
example of this methodology is given in review DR23
14
. One issue with this approach is that
it can be a poor guide to the presence or absence of HFT. For example, in DR23, the trading
venues with the highest message trafc were in China, where HFT penetration is relatively light
compared with London or Frankfurt. The message trafc in the Chinese case is generated by a
large number of lower frequency retail traders.
2. To use (where available) trader identities which allows for classication through their observed
strategies. Examples using US data can be found in Brogaard (2010)
15
and Hendershott et al.
( 2011)
16
. In Europe, Menkveld (2012) studied data from Chi-X/Euronext and identied an HFT
market maker by their trading strategy. The driver review DR21
17
uses the Financial Services
Authority (FSA) database and also involves directly classifying traders based on their observed
behaviour.
12 Krugman (2009).
13 The development of the innovations discussed here occurred largely in the past decade, a time period also characterised by
a very large nancial and banking crisis and now a sovereign debt crisis. Moreover, both Europe and the USA saw dramatic
regulatory change in the guise of MiFID and Reg NMS, respectively. Consequently, care must be taken before ascribing most
of the change in the market’s behaviour to particular technological innovations.
14 DR23 (Annex D refers).
15 Brogaard (2010).
16 Hendershott et al. (2011).
17 DR21 (Annex D refers).
44
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
Measurement of market quality
A second issue concerns how to measure market quality: namely liquidity, price efciency and
transaction costs. Some of the evidence reviewed in this Report has used quoted bid-ask spreads
and depth at the best quote, which are based solely on the order book. These methods have
been criticised by some as representing a mirage of liquidity
18
. They also do not take account of
market impact, which is an important component to transaction costs. However, other evidence
is also reviewed where measures are based on effective and realised spreads, which make use
of the order book and the transaction record, taking into account the impact of a trade. This
counters the mirage problem to some extent. These approaches are widely used by academics
and practitioners. The SEC mandates that trading venues in the US report and disseminate these
measures monthly through their Rule 605
19
. This rule states:
The term average effective spread shall mean the share-weighted average of effective
spreads for order executions calculated, for buy orders, as double the amount of
difference between the execution price and the midpoint of the consolidated best
bid and offer at the time of order receipt and, for sell orders, as double the amount
of difference between the midpoint of the consolidated best bid and offer at the time
of order receipt and the execution price.
This impounds price movements after the trade (including the price impact due to the
information in the trade). This cost can be interpreted as the prot realised by the other side of
the trade, assuming they could lay off the position at the new quote midpoint. The realised spread
and effective spread can be combined to measure the price impact, specically:
Price impact = (effective spread – realised spread) / 2
Other liquidity measures include daily traded volume and turnover. The so-called Amihud
measure of illiquidity, which measures the amount of price change per unit of trading volume, is
also widely used and is described in review DR1
20
.
Variance ratios are standard ways to measure price efciency and are discussed in the review
DR12
21
. They compare, for example, the variance of weekly returns with (ve times) the variance
of daily returns. If markets were fully efcient this ratio should be one, and any departure from
this is evidence of market inefciency. Low variance ratios imply negative dependence, meaning
that prices tend to overreact, whereas high variance ratios imply positive dependence, meaning
that prices tend to follow short-term trends.
18 Haldane (2011).
19 See Hasbrouck (2010).
20 DR1 (Annex D refers).
21 DR12 (Annex D refers).
45
The Future of Computer Trading in Financial Markets
Statistical issues
A key issue in identifying the consequences of HFT for market quality is endogeneity. That is,
property x may be both a cause of HFT activity and a consequence of HFT activity. For example,
HFT may cause some volatility, but some volatility may cause HFT to increase since this offers
many protable trading opportunities. The econometric methods available to identify the
direction of causation or rather to control for endogeneity are as follows.
In a panel data context, where the variable of interest and the amount of HFT per unit per day
is measured, differences in differences may be taken; that is, the difference between treated and
untreated before and after the treatment (where treatment means the active presence of HFT)
are compared. This approach eliminates common time-specic and rm-specic effects that may
contaminate our view of the effects. This approach allows for a limited type of endogeneity but is
very simple to apply and widely used in applied microeconomics.
A second approach is to use an instrumental variable; that is, a variable that is related to the input
variable (for example, HFT activity) but unrelated to the output variable (for example, market
quality) except directly through the input variable HFT. For example, the review DR21 used
latency upgrade events on the London Stock Exchange (LSE). Analysis in review DR23
22
used
the time at which an exchange rst adopted automation. The main problem with this approach is
nding credible instruments, that is, those that are not related directly to the output variable.
3.3.1 Liquidity
Liquidity is a fundamental property of a well functioning market, and lack of liquidity is generally at the
heart of many nancial crises and disasters. Dening liquidity, however, is problematic. At its simplest
level, a market is liquid if a trader can buy or sell without greatly affecting the price of the asset. This
simple statement, however, raises a variety of questions. Does it matter how much a trader wants to
trade? Does time, or how long it takes to execute a trade, also matter? Will liquidity depend in part on
the trading strategy employed? Might not liquidity mean different things to different traders?
Academics and market practitioners have developed a variety of approaches to measure liquidity.
Academics argue that liquidity is best measured or dened by attributes such as tightness, resilience
and depth
23
. Tightness is the difference between trade price and original price. Depth corresponds
to the volume that can be traded at the current price level. Resilience refers to the speed with which
the price returns to its original level after some (random) transactions. In practice, researchers and
practitioners rely on a variety of measures to capture liquidity. These include quoted bid-ask spreads
(tightness), the number of orders resting on the order book (depth) and the price impact of trades
(resilience). These order book measures may not provide a complete picture since trades may not take
place at quoted prices, and so empirical work considers additional measures that take account of both
the order book and the transaction record. These include the so-called effective spreads and realised
spreads, which are now widely accepted and used measures of actual liquidity
24
. Most of the studies
reported below use these metrics.
In current equity markets, a common complaint is that liquidity is transient, meaning that orders are
placed and cancelled within a very short time period and so are not available to the average investor.
For example, the survey commissioned by this Project
25
reports that institutional buy-side investors
believe that it is becoming increasingly difcult to access liquidity and that this is partly due to its
46
22 DR23 (Annex D refers).
23 Kyle (1985) and O’Hara (1995) for discussion.
24 In the USA, information about effective spreads must be provided by market centres that trade national market system
securities on a monthly basis as a result of Exchange Act Rule 11Ac1-5.
25 See Survey of end-users in SR1 (Annex D refers).
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
fragmentation on different trading venues, the growth of ‘dark’ liquidity
26
and the activities of high
frequency traders. The counterpoint to this argument is that algorithms now split large orders (the
so-called ‘parent’ order) into smaller ‘child’ orders that are executed over time and location. Like the
unseen parent order, these child orders are often not totally displayed to the market. So liquidity per se
is more of a moving target, and traders seek it out using various computer-driven strategies. A variety
of algorithms, such as Credit Suisse’s ‘Guerrilla’, Goldman Sachs’ ‘Stealth’, or ITG’s ‘Dark’, are designed
to nd liquidity without revealing the trading intentions, or even the existence, of the order submitter.
These complications are not without costs to investors: the commissioned survey SR1 reports buy-
side concerns that in order to access liquidity they have had to ‘tool up’ on expensive hardware and
software or become more reliant on broker solutions that embed these costs.
The main empirical question is whether CBT (either in the guise of algorithmic trading (AT) or high
frequency activity more generally) is associated with a decrease or increase in liquidity during regular
market conditions. An equally important question relates to whether such trading exacerbates liquidity
problems in situations of market stress.
There are a number of studies that try to identify CBT and its consequences on the order book and
transactions. Hendershott et al. (2011) use the automation of the NYSE quote dissemination as an
implicit experiment to measure the causal effect of AT on liquidity
27
. In 2003, the NYSE began to phase
in the auto-quote system, which empowered CBT, initially for six large active stocks and then slowly
over the next ve months to all stocks on NYSE. They found that this change narrowed spreads, which
was interpreted as increased AT improving liquidity and reducing adverse selection. The evidence was
strongest for large stocks. Another study by Chaboud et al. (2009) also reports results on liquidity
in the Electronic Broking Services (EBS) exchange rate market
28
. They found that, even though some
algorithmic traders appear to restrict their activity in the minute following macroeconomic data
releases, they increased their supply of liquidity over the hour following each release.
Hasbrouck and Saar (2010) investigate order book data from NASDAQ during the trading months of
October 2007 and June 2008
29
. Looking at 500 of the largest rms, they construct a measure of HFT
activity by identifying ‘strategic runs’, which are linked submissions, cancellations and executions. These
are likely to be parts of a dynamic strategy adopted by such traders. Their conclusion is that increased
low-latency activity improves traditional market quality measures such as spreads and displayed depth
in the limit order book, as well as reducing short-term volatility.
Brogaard (2010) also investigated the impact of HFT on market quality in US markets
30
. High frequency
traders were found to participate in 77% of all trades and tended to engage in a price-reversal strategy.
There was no evidence to suggest that high frequency traders were withdrawing from markets in
bad times or engaging in abnormal front-running of large non-HFT trades. High frequency traders
demanded liquidity for 50.4% of all trades and supplied liquidity for 51.4% of all trades. They also
provided the best quotes approximately 50% of the time.
Turning to Europe, Menkveld (2012) studied in some detail the entry of a new high frequency trader
into trading on Dutch stocks at Euronext and a new market Chi-X in 2007 and 2008
31
. He shows that
the inventory of the high frequency trader ends the day close to zero but varies throughout the day,
which is consistent with the SEC denition of HFT. All the trader’s earnings arose from passive orders
(liquidity supply). He also found that the bid-ask spreads were reduced by about 30% within a year
when compared with Belgian stocks that were not traded by the HFT entrant.
26 Regarding the FTSE All-Share index for the week beginning 6th August, 2012: 47.9% of trading volume occurred on lit venues,
46.1% occurred on OTC venues, 3.8% was reported on dark pools, and 2.3% on Systematic Internalizers. See FIDESSA
http://fragmentation.dessa.com/fragulator/ The categories: lit, OTC, dark, and SI are dened by MiFID and they contain some
overlap, but the only category that is fully transparent is the lit class. Even the lit class of venues such as the LSE allows so-
called iceberg orders that are only partially revealed to the public. Accessed: 17 September 2012.
27 Hendershott et al. (2011), op. cit.
28 Chaboud et al. (2009).
29 Hasbrouck & Saar (2011).
30 Brogaard (2010).
31 Menkveld (2012).
47
The Future of Computer Trading in Financial Markets
There are also studies reporting trends in liquidity without specically linking them to algorithmic or
HFT. Castura et al. (2010) investigated trends in bid-ask spreads on the Russell 1000 and 2000 stocks
over the period 2006 to 2010
32
. They show that bid-ask spreads have declined over this period and
that available liquidity (dened as the value available to buy and sell at the inside bid and ask) improved
over time. Angel et al. (2010) show a slow decrease in the average spread for S&P500 stocks over
the period 2003 to 2010 (subject to some short-term up-side uctuations in 2007/2008)
33
. They also
nd that depth has increased slowly over the relevant period. The evidence also shows that both the
number of quotes per minute and the cancellation-to-execution ratio have increased, while market
order execution speed has increased considerably.
Two commissioned studies provide evidence for UK markets. Friedrich and Payne (DR5) compare the
operation of HFT in equities and FX. They nd that penetration of algorithmic, dynamic agency ow
(i.e. best execution of trades on behalf of clients) on multilateral order books in FX is small relative to
equities, perhaps because FX is more liquid and therefore orders do not have to be broken up. They
report no trend in volume (the traded value) of FTSE100 stocks traded between 2006 and 2011, but
nd that bid-ask spreads have decreased while depth has increased. The number of trades, on the
other hand, has increased more than vefold over this period, implying that the average trade size is
now only 20% of its former level. For small UK stocks, there are different results. First, the average
trade size has not changed as much over the period 2006 to 2011, which suggests that HFT is not
so actively involved in their trading. Second, there has been little improvement in the liquidity of small
cap stocks.
Linton (DR1) measures the daily illiquidity of the FTSE All-Share index using a low-frequency measure
– the absolute return to unit of volume
34
. He nds this measure of illiquidity has varied considerably
over the past ten years, rst declining then rising during the nancial crisis before falling again. Figure 3.1
shows an updated plot of the series up to June 2012.
48
32 Castura et al. (2010).
33 Angel et al. (2010).
34 The FTSE All-Share Index, (originally known as the FTSE Actuaries All Share Index,) is a capitalisation-weighted index,
comprising around 1,000 of more than 2,000 companies traded on the London Stock Exchange. As at June 2011, the
constituents of this index totalled 627 companies (source: FTSE All-Share fact sheet). It aims to represent at least 98% of the
full capital value of all UK companies that qualify as eligible for inclusion. The index base date is 10 April 1962, with a base level
of 100. The Amihud illiquidity measure is described in Box 3.1; it measures percentage price change per unit of trading volume.
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
Figure 3.1: FTSE All-Share illiquidity 1997-2012
Illiquidity (times 10*11)
5.0
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
1998 2000 2002 2004 2006 2008 2010 2011 2012
Year
Source: FTSE
The same is true of traded volume. The process driving traded volume of UK equities seems to be
highly persistent, which means that bad shocks to volume, like that which occurred in 2008/2009, can
take a long time to correct. The main conclusion from this study is that the macroeconomic crises of
the past ve years have been the main driving force for the liquidity and volatility of UK equity markets.
If HFT has a role to play in the large swings in market conditions, it is relatively small and insignicant
in comparison with the huge negative effects of the banking and sovereign debt crises that happened
during this period. The level of the FTSE All-Share index has oscillated wildly throughout the past
decade, as can be seen from Figure 3.2. Any medium term investor who was in the market during this
period would have faced the prospect of having the value of his investments cut in half on two separate
occasions and, if he held the market throughout the whole period, he would have ended up more or
less where he started in nominal terms.
49
The Future of Computer Trading in Financial Markets
Figure 3.2: FTSE All-Share Index 1997-2012
3800
3400
3000
Price level
2600
2200
1800
1400
1998 2000 2002 2004 2006 2008 2010 2011 2012
Year
Source: FTSE
The fragmented nature of liquidity was remarked on by a number of contributors to the survey of end
users commissioned by the Project and reported in SR1
35
. In Europe, this fragmentation was facilitated
by the Markets in Financial Instruments Directive (MiFID) regulation of November 2007, which
permitted competition between trading venues. This market structure has been a seedbed for HFT,
which has beneted from the competition between venues through the types of orders permitted,
smaller tick sizes, latency and other system improvements, as well as lower fees and, in particular, the
so-called maker-taker rebates. It is almost impossible to access liquidity effectively on multiple venues
without sophisticated computer technology and, in particular, smart order routers (SOR).
35 SR1 (Annex D refers).
50
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
There are a number of studies that provide evidence on the effects of competition between trading
venues. O’Hara and Ye (2011) nd that stocks with more fragmented trading in the USA had lower
spreads and faster execution times
36
. Gresse (DR19)
37
investigates whether this liquidity fragmentation
has had a positive or negative effect on market quality in Europe. She examines this from two points of
view. First, from the perspective of a sophisticated investor who has access to SORs and can access the
consolidated order book. Second, from the point of view of an investor who can only access liquidity
on the primary exchange. She compares FTSE100, CAC40 (French large caps) and SBF 120 (French
mid caps) stocks across pre-and-post MiFID periods. She nds that fragmentation has generally had
positive effects on spreads both locally and globally, especially for the larger cap stocks, whereas for the
smaller cap SB120 the improvements were minor. For example, from October 2007 to September 2009,
average (global) spreads on FTSE100 stocks fell from 9.21 to 5.43 basis points (bps), while the local
spreads on average fell to 7.07, somewhat less. Spreads narrowed far less dramatically for the mid caps
of the SBF 120.
These summary statistics are supplemented by a panel data analysis that controls for other variables
that might affect outcomes and, in particular, acknowledges the background of the nancial crisis and
its potential effect on market quality. The regression study generally conrms the ndings, although the
improvements of the SBF 120 spreads are not statistically signicant. This is consistent with the fact that
there is much less competition in liquidity supply for these stocks.
Some issues that arise in a fragmented marketplace are locked and crossed markets, and trade-throughs.
A trade-through occurs when a trade is executed on one venue at a price that is inferior to what was
available at another venue at the same time (this is, in principle, not permitted in the USA through the
Regulation National Market System (NMS), although such occurrences are not prohibited in Europe
where best execution is dened along more dimensions than price alone). A locked (crossed) market
occurs where the bid-ask spread is zero (negative) and is evidence of poor linkages between markets
and investors. In fact, such market failures are relatively infrequent and there is some evidence that they
have decreased in the past few years
38
, presumably through the activities of HFT arbitrageurs.
Some have argued that maker-taker pricing exacerbates liquidity issues in a fragmented marketplace.
The commissioned study by Foucault (DR18) considers this question. Under this exchange-pricing
structure, successful liquidity suppliers are given rebates. The stated aim is to encourage liquidity supply.
Since high frequency traders are better able to place their limit orders at the top of the queue, they
are the prime direct beneciaries. Nevertheless, Foucault argues that if routing decisions are based
on quotes cum fees and there is no minimum tick size, then the fee structure should be irrelevant
to market outcomes. In reality these conditions do not hold: in particular, routing decisions taken by
agents of traders are often not based solely on the total fee. Furthermore, tick sizes are not zero and
in some cases are imposing binding constraints. He argues that maker-taker pricing in this environment
may change the balance of aggressive market orders and passive limit orders and could have a
benecial effect on aggregate welfare.
Unfortunately, there is not much evidence on the effects of this pricing structure. The one empirical
study of this issue, which is based on an experiment conducted on the Toronto Stock Exchange, suggests
little negative impact of the maker-taker fee structure. The conclusion of review DR18 is that more
empirical work is needed to allocate the costs and benets of this market structure properly before
policy options are exercised. These issues are examined in greater detail in Chapter 6 of this report.
36 O’Hara & Ye (2011).
37 DR19 (Annex D refers).
38 Ende & Lutat (2010).
51
The Future of Computer Trading in Financial Markets
In summary, most evidence points to improvement in liquidity in nancial markets as a result of CBT
and HFT. However, there may be some issues to do with liquidity provision in times of market stress.
This general conclusion has been supported by other government-commissioned studies, including
those by the Australian, Canadian and Swedish national regulators
39
.
3.3.2 Transaction costs
Trading with computers is cheaper than trading with humans, so transaction costs have fallen steadily
in recent years as a result of the automation of markets. Jones (2002) reports the average relative
one-way costs paid for trading Dow Jones stocks between 1935 and 2000
40
. He nds the total cost of
trading has fallen dramatically in the period 1975 to 2000. Angel et al. (2010) show that average retail
commissions in the USA have decreased between 2003 and 2010, a period more relevant for inferring
the effects of computer trading. They also make a cross-country comparison of trading costs at the end
of 2009. According to this study, USA large cap stocks are the cheapest to trade in the world with a
roughly 40 basis point cost. Incidentally, the UK fared quite poorly in this comparison, with an average
90 basis point cost that was worse than the rest of Europe and Canada and only marginally better than
emerging economy stocks.
Menkveld (DR16)
41
argues that new entry, often designed to accommodate HFT, had profound effects
on transaction costs. For example, the entry of Chi-X into the market for Dutch index stocks had an
immediate and substantial effect on trading fees for investors, rst through the lower fees that Chi-X
charged and then through the consequent reduction in fees that Euronext offered. The strongest
effect, however, was a reduction in clearing fees. A new clearing house entered, EMCF, and this
triggered a price war that ended up with a 50% reduction in clearing fees.
This reduction in clearing fees seems to have been replicated across European exchanges to the
benet of investors. Unfortunately, settlement fees, which are perhaps an even larger component
of transaction costs, remain uncontested in Europe and have not fallen to the lower levels benetting
US investors.
Although most academic studies have shown that HFT participation has improved market quality
measures such as spreads (which reect some transaction costs), they have not directly addressed the
question of whether long-term investors have been positively or negatively affected. It could be that
they are targeted by high frequency traders through their speed advantage and through predatory
strategies and that the main beneciaries of the improved metrics are the high frequency traders
themselves. The commissioned survey SR1 provides some support for this contention: while it showed
that institutional investors generally accept that commissions and bid-ask spreads have decreased as a
result of the market fragmentation and HFT facilitation of MiFID, they have concerns that most of the
39 These include:
a) The US CFTC technical advisory committee meeting http://www.cftc.gov/ucm/groups/public/@newsroom/documents/
le/tac_032912_transcript.pdf Accessed: 17 September 2012.
b) The IOSCO report http://www.iosco.org/library/pubdocs/pdf/IOSCOPD354.pdf p. 26, l2: “The available evidence fails to
nd a consistent and signicant negative effect of HFT on liquidity” Accessed: 18 September 2012.
c) Bank of Canada http://www.bankofcanada.ca/wp-content/uploads/2011/12/fsr-0611-barker.pdf p. 48. : “HFT appears to be
having a profound impact on market liquidity, and its rise has coincided with an increase in trading volumes, tighter bid-
offer spreads and lower market volatility, at least during periods when markets are not experiencing stress.”) Accessed:
17 September 2012.
d) Swedish FSA http://www.mondovisione.com/media-and-resources/news/swedish-nancial-supervisory-authority-
nansinspektionen-high frequency-tra/, “ investigation has demonstrated that the impact of HFT on trading is smaller
than feared. Swedish investors believe that trading has undergone a transformation and that the market has become
more volatile, but that these changes can be explained by multiple factors and not only the emergence of HFT” Accessed:
18 September 2012.
e) Australian Regulators http://www.asic.gov.au/asic/pdib.nsf/LookupByFileName/rep-215.pdf/$le/rep-215.pdf, p. 91:
“The benets to the economy and the equities industry from competition are likely to outweigh the costs of order ow
fragmentation, which include those of increased surveillance, technology and information. The net benet will be positive
if competition is introduced with the proposed rules framework”; p. 94: “Some current industry players will thrive in the
new conditions, while others will nd the environment extremely challenging. Those with the technology and volume
to gain full benet from the multiplicity of execution venues may experience lower costs and new business/trading
opportunities” Accessed: 17 September 2012
f) The LSEG submission on MiFID http://www.londonstockexchange.com/about-the-exchange/regulatory/lseg-submission-
to-ec-on-mid02-02-11.pdf : Regarding AT/HFT, p. 1: “No evidence of market failure” Accessed: 17 September 2012.
40 Jones (2002).
41 DR16 (Annex D refers).
52
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
benets are captured by their brokers. These investors believe that infrastructure costs have increased
as well as some other indirect costs
42
.
The interests of institutional investors are of great importance. A commissioned study by Brogaard
et al. (DR21) examines the direct effects of HFT on the execution costs of long-term investors.
The authors use a new UK dataset obtained from the detailed transaction reports of the UK Financial
Services Authority (FSA) over the period 2007 to 2011 to provide a better measurement of HFT
activity. They combine this with Ancerno data on institutional investors’ trading costs. To test whether
HFT has impacted the execution costs of institutional traders, the authors conduct a series of event
studies around changes in network speeds on the LSE to isolate sudden increases in HFT activity.
This study found that the increases in HFT activity have no measurable effect on institutional execution
costs. Of course, additional studies linking HFT and institutional trading costs in other market settings
would be helpful in determining the generality of this nding.
In summary, the evidence is that transaction costs have declined in the past decade, mostly due to
changes in trading market structure (which are closely related to the development of HFT). Specically,
there has been increased competition between trading venues and greater use of labour-saving
technology.
3.3.3 Price efficiency
A central claim of nancial economists is that more efcient prices (which better reect fundamental
values) in nancial markets contribute to more informed nancing and investment decisions and,
ultimately, to better allocation of resources in the wider economy and, hence, higher welfare. The
holy grail of fully efcient markets is now widely acknowledged to be impossible, following the work of
Grossman & Stiglitz (1980)
43
: if obtaining information and trading on it is costly, then no one will gather
it unless they can prot, but if all the information is already in prices, then no one can prot from it. If
no one gathers information, then how can it be impounded in prices? In the light of this, the relevant
question is how inefcient markets are and with respect to which information.
Hendershott (DR12) describes the meaning of price efciency in the context of high-speed markets,
and presents the arguments why HFT may improve market efciency by enabling price discovery
through information dissemination. The usual method of measuring the degree of market inefciency
is through the predictability of prices based on past price information alone. In practice, widely used
measures such as variance ratios and autocorrelation coefcients estimate the predictability of prices
based on linear rules.
Several studies commissioned by this Project address the question of whether HFT strategies are likely
to improve or worsen price efciency. Brogaard (DR10) describes how high frequency traders make
money and how their activities may affect price discovery (for example, by making the prices of assets
with similar payoffs move more closely together). The prots of HFT come from a variety of sources
including passive market making activity, liquidity rebates given by exchanges to reward liquidity supply
and statistical pattern detection used in so-called stat-arb strategies. Greater market making activity
should improve efciency. HFT strategies that enforce the law of one price across assets and across
trading venues similarly improve the quality of prices facing investors. Farmer (DR6) cautions that, as
market ecologies change, the transition to greater efciency may be slow.
42 One reason for increased transaction costs for large traders may be quote matching. Suppose that a large trader places a limit
order to buy at 30. A clever trader who sees this order could immediately try to buy ahead of it, perhaps by placing an order
at 30 at another exchange, or by placing an order at a tick better at the same exchange. If the clever trader’s order lls, the
clever trader will have a valuable position in the market. If prices subsequently rise, the trader will prot to the extent of the
rise. But if values appear to be falling, perhaps because the prices of correlated stocks or indices are falling, the clever trader
will try to sell to the large trader at 30. If the clever trader can trade faster than the large trader can revise or cancel his order,
and faster than can other traders competing to ll the large trader’s order, then the clever trader can limit his losses. The
clever trader thus prots if prices rise, but loses little otherwise. The large trader has the opposite position: if prices rise,
he may fail to trade and wish that he had. If prices fall, he may trade and wish that he had not. The prots that the clever
trader makes are lost prot opportunities to the large trader. The quote-matching strategy was established long before HFT,
however. See Harris (1997) Order Exposure and Parasitic Traders.
43 Grossman & Stiglitz (1980).
53
The Future of Computer Trading in Financial Markets
Brogaard et al. (2012)
44
nd that high frequency traders play a positive role in price efciency by
trading in the direction of permanent price changes and in the opposite direction of transitory pricing
errors on average days and the days of highest volatility. This is done through their marketable orders.
In contrast, HFT passive orders (which are limit orders that are not immediately marketable) are
adversely selected in terms of the permanent and transitory components as these trades are in the
direction opposite to permanent price changes and in the same direction as transitory pricing errors.
HFT marketable orders’ informational advantage is sufcient to overcome the bid-ask spread and
trading fees to generate positive trading revenues. Non-marketable limit orders also result in positive
revenues as the costs associated with adverse selection are smaller than the bid-ask spread and
liquidity rebates.
Negative effects on efciency can arise if high frequency traders pursue market manipulation strategies.
Strategies such as front running, quote stufng (placing and then immediately cancelling orders) and
layering (using hidden orders on one side and visible orders on the other) can be used to manipulate
prices. For example, deterministic algorithmic trading such as volume weighted average price (VWAP)
strategies can be front run by other algorithms programmed to recognise such trading. Momentum
ignition strategies, which essentially induce algorithms to compete with other algorithms, can push
prices away from fundamental values. However, it is clear that price efciency-reducing strategies, such
as manipulative directional strategies, are more difcult to implement effectively if there are many rms
following the same strategies. Thus, the more competitive the HFT industry, the more efcient will be
the markets in which they work.
There is a variety of evidence suggesting that price efciency has generally improved with the growth
of CBT. Castura et al. (2010) investigate trends in market efciency in Russell 1000/2000 stocks over
the period 1 January 2006 to 31 December 2009
45
. Based on evidence from intraday
46
variance ratios,
they argue that markets become more efcient in the presence of and increasing penetration by HFT.
Johnson and Zhao (DR27) look at ultra-high frequency data and nd some evidence that (relatively)
large price movements occur very frequently at the machine time-scale of less than half a second.
Brogaard (2010) provides further evidence on this issue. He estimates that the 26 HFT rms in his
sample earn approximately $3 billion in prots annually. Were high frequency traders not part of the
market, he estimates that a trade of 100 shares would result in a price movement of $.013 more than it
currently does, while a trade of 1,000 shares would cause the price to move by an additional $.056. He
also shows that HFT trades and quotes contribute more to price discovery than does non-HFT activity.
Linton (DR1) provides evidence based on daily UK equity data (FTSE All-Share). Specically, he
computes variance ratio tests and measures of linear predictability for each year from 2000 to
2010. The measures of predictability (inefciency) uctuate around zero with sometimes more and
sometimes less statistical signicance. During the nancial crisis there was somewhat more inefciency,
but this declined (until late 2011 when there was a series of bad events in the market). Overall, DR1
found no trend in daily market efciency in the UK market, whether good or bad. Figure 3.3 shows an
updated graph showing the variance ratio of the FTSE All-Share index until June 2012
47
.
44 Brogaard et al. (2012).
45 Castura et al. (2010), op. cit.
46 They look at 10:1 second ratios as well as 60:10 and 600:60 second ratios.
47 This is calculated using a rolling window of 250 daily observations (roughly one trading year). It is the weekly variance ratio
which is the variance of weekly returns divided by ve times the variance of daily returns. If returns were unpredictable (i.e.,
markets were efcient) this ratio should be one, which is displayed by the middle dashed line. The upper and lower dashed
lines represent 95% condence intervals centred at one. There are few instances when the condence bands over this period
are exceeded, with notable exceptions in 2008/2009 and a brief violation in 2011. The daily return series is of interest to many
retail investors.
54
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
Figure 3.3: FTSE All-Share daily return predictability 2000-2012
1.4
Weekly Variance Ratio
1.2
1.3
Upper confidence band
1.1
1.0
0.9
0.7
0.8
Lower confidence band
0.6
2002 2003 2004 2005 2006 2007
Year
2008 2009 2010 2011 2012
Source: FTSE
One further issue is the location of price discovery and whether this has been affected by the increased
competition between venues. This is an important issue in the European context where the few
regulated markets (like the LSE) claim a unique position, based partly on their supposed pivotal role
in price discovery. Gresse (DR19) nds that active multilateral trading facilities (MTF), such as Chi-X in
London, played a signicant part in price discovery. Furthermore, this transfer of price discovery has
not worsened the efciency of the consolidated market in Europe.
These studies have all addressed the functioning of the secondary market for equity trading. Boehmer
et al. (DR23) have investigated the effects of AT on rms’ equity capital-raising activities in an
international panel data-set. They nd some evidence that the yearly AT intensity (which they measure
by the message trafc in a stock) reduces net equity issues over the next year with the main channel
being the effect on share repurchases. It is not clear what the mechanism is for this effect, so these
statistical results should be treated with some caution.
55
The Future of Computer Trading in Financial Markets
In summary, the preponderance of evidence suggests that HFT has not harmed, and may have
improved, price efciency. It may also follow from this that HFT has improved welfare, although the
connection between the two concepts is complex
48
.
3.4 What has been the impact of computer-based trading on market quality?
The Flash Crash in US markets has brought increased scrutiny to the role of episodic illiquidity in
markets and its relation to the current computer-based market structure. The events of 6 May
2010 have now been extensively documented in two reports by the Commodity Futures Trading
Commission (CFTC) and SEC staff. These reports show that a complex interaction of forces led to the
Dow Jones Industrial Average falling 998 points, the largest intraday drop in US market history. While
the Flash Crash lasted less than 30 minutes, for a brief interval more than one trillion dollars in market
capitalisation was lost. In the aftermath of the crash, more than 20,000 trades were cancelled. A more
lasting effect has been a sustained withdrawal from equity investing by retail traders.
The CFTC and SEC reports highlight the role played by a large algorithmic sell trade in the S&P e-mini
futures contract that coincided with the beginning of the crash. While clearly an important factor, the
reports also highlight the roles played by a variety of factors, such as routing rules, quoting conventions,
internalisers (the name given to banks and broker/dealer rms that clear order ow internally), high
frequency traders and trading halts. These reports make clear two compelling facts about the current
market structure: episodic illiquidity can arise, and when it does, it is rapidly transmitted to correlated
markets. That the Flash Crash began in what is generally viewed as one of the most liquid futures
contracts in the world only underscores the potential fragility of the current market structure.
A variety of research considers how CBT may be a factor in precipitating periodic illiquidity. Leland
(DR9) highlights the role that forced selling can have on market liquidity. Such selling can arise from
various trading strategies and its effects are exacerbated by leverage. Leland argues that AT strategies
are triggered in response to automatic data feeds and so have the potential to lead to cascades in
market prices as selling causes price falls which prompt additional selling. Leland also argues that, due
to forced selling, the crash of 19 October 1987 and the Flash Crash have great similarities. As modern
HFT did not exist in 1987, this result underlines that market illiquidity is not a new event. What
matters for this Project is whether current computer-driven practices are causing greater illiquidity risk.
Madhavan (2011) argues that fragmentation linked to HFT may be one such cause
49
. He shows that
fragmentation measured using quote changes, which he argues is reective of HFT activity, has high
explanatory power with respect to cross-sectional effects on equity instruments in the Flash Crash.
Kirilenko et al. (2011) provide a detailed analysis of high frequency futures traders during the Flash
Crash
50
. They found that high frequency traders initially acted as liquidity providers but that, as prices
crashed, some high frequency traders withdrew from the market while others turned into liquidity
demanders. They conclude that high frequency traders did not trigger the Flash Crash but their
responses to unusually large selling pressure increased market volatility. Easley et al. (2011) argue that
historically high levels of order toxicity forced market makers to withdraw during the Flash Crash
51
.
Order ow is considered toxic when it adversely selects market makers who are unaware that they
are providing liquidity at their own loss. Easley et al. (2011a) develop a metric to measure such toxicity
and argue that order ow was becoming increasingly toxic in the hours leading up to the Flash Crash.
48 Jovanovic & Menkveld (2011) provide a model that formalises the role that high frequency traders perform as middlemen
between ‘real’ buyers and ‘real’ sellers. They compare a market with middlemen to a market without. Their model allows
for both positive and negative effects (of middlemen) depending on the value of certain parameters that measure the
informational advantage that high frequency traders would have through their ability to process numerical information about
the order book rapidly. They calibrate their model using the empirical work presented in Menkveld (2012) about the entry of
an HFT market maker on Dutch index stocks on Chi-X. Empirically, HFT entry coincided with a 23% drop in adverse-selection
cost on price quotes and a 17% increase in trade frequency. They conclude that, overall, there is a modest increase in welfare.
49 Madhavan (2011).
50 Kirilenko et al. ( 2011).
51 Easley et al. ( 2011a).
56
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
Other factors were at play during the Flash Crash; for example, the delayed and inaccurate data feeds
and the breaking of the national market system through self-help declarations by various exchanges
52
.
There have been a variety of other, smaller illiquidity events in markets since the Flash Crash. On 8
June 2011, for example, natural gas futures plummeted 8.1% and then bounced back seconds later. On
2 February 2011, an errant algorithm in oil futures sent 2,000–3,000 orders in a single second, causing
an eightfold spike in volatility and moving the oil price $1 before the algorithm was shut down. In
March, trades in ten new Morningstar exchange traded funds (ETF) were cancelled when prices fell by
as much as 98% following what was determined to be a ‘fat-nger problem’ (the colloquial name given
to an error made while entering data). On 24 August 2010, there were very rapid changes in the prices
of ve LSE-listed stocks: BT Group PLC, Hays PLC, Next PLC, Northumbrian Water Group PLC and
United Utilities Group PLC. In this case, there appear to have been no signicant news about macro-
economic or stock-specic fundamentals. The rapid falls triggered the LSE circuit breakers, which
effectively prevented further falls and contagion across the broader market. Chapter 6 discusses the
efcacy of the existing market controls.
There are also many cases where prices have moved substantially within a day driven by fundamental
news rather than by trading system issues. The 1929 stock market crash is widely believed to be a
correction of previous overvaluations, and the 1987 crash is also believed to be at least partly driven
by fundamentals, since it was echoed around the world. Many rm-specic crashes have occurred that
were due simply to bad news about the performance prospects of the companies involved
53
.
In summary, the evidence suggests that HFT and AT may be contributing to periodic illiquidity in
current markets. They may not be the direct cause of these market crises, but their trading methods
mean that their actions or inactions may temporarily amplify some liquidity issues.
3.5 Future: how is the impact of computer-based trading on liquidity likely to
evolve in the next ten years?
There is considerable uncertainty regarding the future role of HFT. TABB Group (2012) estimates
that HFT accounted for 53% of trading in US markets in the rst half of 2011, a decrease from the
61% share it held in 2010
54
. However, the extreme market volatility in August 2011 saw HFT return
with a vengeance. Wedbush, one of the largest providers of clearing services to high frequency rms,
estimates that HFT made up 75% or more of US volume between 4 and 10 August 2011
55
. Whether
HFT will continue to be as active when volatility subsides to normal levels remains to be seen
56
.
There are also indications that the protability of HFT is reaching its limits and in the next ten years
may come under further pressure. Such reductions may arise for a variety of reasons: the potential
move to sub-penny pricing in the USA may reduce protability from market making, new types of
multi-venue limit orders may be permitted that will reduce the potential for stale prices across different
trading venues, new entrants to the HFT industry will take prots from incumbents and regulation and
taxation may make their business model unviable. Lowering costs of entry, which may arise from future
technological improvements, can also improve competition. Limiting the value of small improvements in
speed by, for example, reducing the value of time priority or requiring a minimum quote life, may also
reduce HFT, because it will reduce the incentives for a winner-take-all speed race.
52 See http://www.sec.gov/divisions/marketreg/rule611faq.pdf for an explanation of the self help exception the NMS rules.
Essentially, exchanges are allowed in exceptional circumstances (such as those occurring on May 6th, 2010) to not exchange
messages with other exchanges, which would be their obligation under Regulation NMS. Accessed: 17 September 2012.
53 For a recent example, Thomas Cook (a FTSE 250 company at the time) closed on 21 November 2011 at 41.11 and closed
at 10.2 on 22 November 2011. The 75% loss of value was attributed to an announcement of bad results that took place
before trading began. Most of the value loss was present at the opening of the market on the 22 November but the decline
continued rapidly through the rest of the day (to the extent that it triggered a number of trading halts). There is some
perception that this loss of value took place at a much faster pace than it would have done ten years ago.
54 TABB Group (2012).
55 Mehta (2011).
56 The increased activity of HFT during this volatile period seems at odds with some earlier arguments that their liquidity
provision can reduce during times of market stress. Perhaps this illustrates the difference between regular volatility, which has
both upside and downside price movements, and a market fall where only downside price changes occur.
57
The Future of Computer Trading in Financial Markets
Nonetheless, it seems inevitable that CBT will remain a dominant force in markets over the next ten
years. One reason for this will be technological advances that facilitate the automated extraction,
aggregation and ltering of news
57
. Such news analytics, discussed in Chapter 2, could be used in model
construction for high frequency traders, as well as for portfolio managers. News analytics technologies
currently allow for the electronic ‘tagging’ of news events, corporate lings and the like, allowing
traders with access to this computer technology the ability to see more information faster. Tying such
information into CBT strategies provides a means for traders to capitalise on information before it is
incorporated into market prices. High frequency traders will be well positioned to take advantage of
such nascent technology.
To the extent that such trading puts information into prices more rapidly, markets will benet by
becoming more efcient. However, such strategies also serve to increase the ‘arms race’ in markets
by bestowing greater rewards on the most technologically advanced traders. Consequently, it may
be that all trading evolves towards CBT, reecting the fact that technology diffuses across populations
given time.
As this happens, market systems may experience unwanted negative effects. One such effect is already
present in the problems of message trafc. Message trafc is the name given to computer instructions
to place, change and cancel orders. On any trading day, message trafc far exceeds trading volume, as
many more orders are cancelled or changed than are ever executed. On volatile days, message trafc
can cause market outages due to the inability of servers and other computer components of trading to
handle the ow. Such outages were widespread during the Flash Crash of 6 May 2010. They recurred
in early August 2011, when the extreme volume and volatility took out trading platforms at Goldman
Sachs and other large trading rms in the USA. When this occurs, market liquidity is affected
58
.
A related systematic risk can arise if a large sub-set of market participants are following the same
strategies. For example, if news analytics becomes a driving force in portfolio management, then
sequences of sell (or buy) orders may arrive at the market, all driven by the same information. For
market makers, such orders are ‘toxic’, because the market maker will be acting as counterparty to
agents with better information. As seen in the Flash Crash, when toxicity overwhelms market makers
their strategy is to withdraw, and illiquidity results. Consequently, new risk management products will
need to evolve to allow market makers, traders and regulators to be able to function
59
. The future of
CBT may thus involve more technology capable of controlling the technology controlling the markets.
Finally, it is not possible to legislate market crashes away, any more than it is possible to dispense with
the business cycle. In future crises, regulators need to be able to reconstruct what has happened;
when, where and why. One of the consequences of MiFID was the loosening of the trade reporting
protocols in Europe so that trades can be reported through Trade Reporting Facilities (TRF) other
than the ‘primary’ market. This made ex-post auditing work much more difcult than it is in the USA
where they have had a consolidated tape for some time. Europe is somewhat behind in the collection,
standardisation and analysis of nancial data in comparison to the USA, where the Ofce of Financial
Research (OFR) has been commissioned by the Dodd-Frank Act to found a nancial data centre to
collect, standardise and analyse such data
60
.
58
57 See DR8 (Annex D refers).
58 At the time of writing, many exchanges, including the LSE, have introduced ‘message throttling’ as well as pricing systems which
may mitigate some of these effects.
59 Easley et al. (2011b).
60 See the OFR Annual Report (2012).
The impact of computer-based trading on liquidity, price efciency/discovery and transaction costs
3.6 Conclusions
CBT is now the reality in asset markets. Technology has allowed new participants to enter, new trading
methods to arise and even new market structures to evolve. Much of what has transpired in markets is
for the good: liquidity has been enhanced, transactions costs have been lowered and market efciency
appears to be better, or certainly no worse. The scale of improvements may be fairly small and, in the
short term, they may have been obscured by the background of a very poor performance by
Organisation for Economic Co-operation and Development (OECD) economies and stock market
indexes in particular. However, there are issues with respect to periodic illiquidity, new forms of
manipulation and potential threats to market stability due to errant algorithms or excessive message
trafc that must be addressed. Regulatory changes in practices and policies will be needed to catch up
to the new realities of trading in asset markets. Caution must be exercised to avoid undoing the many
advantages that the high frequency world has brought. Technology will continue to affect asset markets
in the future, particularly as it relates to the ultra-fast processing of news into asset prices.
The next chapter will consider how the increased use of technology in markets affects nancial stability.
59
4 Financial stability and
computer-based trading
Key findings
Economic research to date provides no direct unambiguous
evidence that high frequency computer-based trading has
increased volatility.
However, in specic circumstances, a key type of
mechanism can lead to signicant instability in nancial
markets with computer-based trading: self-reinforcing
feedback loops (the effect of a small change looping back
on itself and triggering a bigger change, which again loops
back and so on) within well-intentioned management and
control processes can amplify internal risks and lead to
undesired interactions and outcomes.
The feedback loops can involve risk management systems
or multiple algorithms, and can be driven by changes in
market volume or volatility, by market news and by delays
in distributing reference data.
A second cause of instability is social: a process known
as ‘normalisation of deviance’, where unexpected and risky
events come to be seen as ever more normal (for example,
extremely rapid crashes), until a disastrous failure occurs.
61
The Future of Computer Trading in Financial Markets
4 Financial stability and
computer-based trading
4.1 Introduction
As described in Chapter 1
1
a broad interpretation of computer-based trading (CBT) is used in this
report. A useful taxonomy of CBT identies four characteristics which can be used to classify CBT
systems. First, CBT systems can trade on an agency basis (i.e. attempting to get the best possible
execution of trades on behalf of clients) or a proprietary basis (i.e. trading using one’s own capital).
Second, CBT systems may adopt liquidity-consuming (aggressive) or liquidity-supplying (passive) trading
styles. Third, they may be classied as engaging in either uninformed or informed trading and fourth,
a CBT algorithm generates the trading strategy or only implements a decision taken by another
market participant.
Much of the current public debate is concerned with the class of aggressive predatory algorithms,
especially those that operate at high speed and with high frequency. Because most nancial institutions’
use of CBT cannot be neatly assigned to only one of the above four categories, it is more fruitful
to think about CBT systems, the algorithms they employ directly and the frequency at which they
trade, rather than to think about the behaviour of a named nancial or trading corporation, such as
a specic investment bank or fund management company. For much the same reasons, the focus of
the discussion in this chapter is not on any one asset class (such as equities, foreign exchange (FX),
commodities or government bonds), but rather on the forces that seem likely to shape future issues of
stability arising from CBT. This chapter summarises the intuition behind some of the more economically
plausible risk factors of CBT: these ‘risk drivers’ can best be viewed as forming the logical basis of
possible future scenarios concerning the stability of the nancial markets. There is no agreed denition
of ‘systemic stability’ and ‘systemic risk’, and the reader is referred to DR29
2
for a discussion and a
survey on empirical systemic risk measures.
4.2 How has computer-based trading affected financial stability in the past?
The raison d’être for nancial markets is to aggregate myriad individual decisions and to facilitate an
efcient allocation of resources in both primary and secondary markets
3
by enabling a timely and
reliable reaping of mutual gains from trade, as well as allowing investors to diversify their holdings. As
with many other aspects of modern life, innovations in technology and in nance allow the repetitive
and numerically intensive tasks to be increasingly automated and delegated to computers. Automation,
and the resulting gains in efciency and time, can lead to benets but can lead also to private and social
costs. The focus of this chapter is solely on possible repercussions of CBT (including high frequency
trading (HFT) in particular) on nancial stability, especially the risks of instability. This should certainly
not be construed as meaning that CBT is socially detrimental or bears only downside risks and costs.
It is hoped that by better understanding the risk-drivers of CBT on nancial stability, the creators,
users and regulators of CBT systems may be able to manage the risks and allow the benets of CBT
to emerge while reducing the social costs.
The conclusions of this chapter apply to any given market structure, but they are especially relevant to
the continuous double auctions of the electronic limit order book that run on traders’ screens in most
of the major nancial markets worldwide. The reason is that even if daily volume is large, the second-
by-second volume may not be. Even in a huge market such as the FX market, a sufciently large order
62
1 See Box 1.1 in Chapter 1, and also DR5 (Annex D refers).
2 DR29 (Annex D refers).
3 When a company issues equities (shares) to raise capital, this is the primary market in action. When the shares are then
subsequently traded among investors and speculators, this is the secondary market in action.
Financial stability and computer-based trading
can temporarily sway prices, depending on how many other orders are in the market (the ‘depth’ of
the market) at the time and how quickly the book is replenished (the ‘resilience’ of the market).
Price volatility is a fundamental measure useful in characterising nancial stability, since wildly volatile
prices are a possible indicator of instabilities in the market and may discourage liquidity provision
4
. In
DR1, Linton notes that fundamental volatility has decreased in the UK equities market since the turmoil
of 2008/2009, and liquidity and trading volume have slowly returned
5
. If HFT contributes to volatility,
Linton argues, it might be expected that the ratio of intraday volatility to overnight volatility would have
increased as HFT became more commonplace, but they nd no evidence to support that hypothesis.
They note that the frequency of large intraday price moves was high during the crisis period, but the
frequency has declined to more normal levels since the end of 2009. However, Boehmer et al. (2012)
6
in a study spanning 39 exchanges in 36 countries have found that higher volatility and CBT activity
move together over the period 2001 to 2009, though causality is not yet clear and the economic
magnitudes appear to be small.
4 Stability can, of course, differ from volatility by placing signicantly greater weight on large, infrequent price changes, especially
if the latter do not appear to be fundamental.
5 See Figure 4.1 for a time series of realised volatility computed as (high-low)/low; the implied volatility index VFTSE follows a
similar pattern.
6 Boehmer et al. (2012).
63
The Future of Computer Trading in Financial Markets
Figure 4.1: FTSE100 volatility between 2000-2012
0
0.02
0.04
0.06
0.08
0.10
0.12
2000 2002 2004 2006 2008 2010 2012
Year
Percentage of volatility
Source: FTSE
64
Financial stability and computer-based trading
CBT and HFT are relatively new phenomena so the empirical literature examining their role is still
nascent. Research thus far provides no direct evidence that HFT has increased volatility
7
, not least
because it is not clear whether HFT increases volatility or whether volatility invites more HFT – or
both or neither
8
. Signicant challenges in evaluating HFT arise because much of its growth coincides
with the 2008/2009 turmoil and data fully characterising HFT are lacking
9
. Indirect studies of CBT and
HFT provide interesting evidence highlighting the importance of further study with better data, but are
subject to various interpretations
10
.
It is something of a cliché to say that CBT can lead to ‘Black Swan’ events, i.e. events that are extremely
rare but of very high consequence when they do occur. Of course, a more computerised world is
more vulnerable to some types of catastrophic events, such as power failures, major solar emissions,
cyber-attacks and outages of server computers. Any one of these could, in principle, lead to system-
wide failure events.
However, the more interesting and signicant aspects of analysing nancial stability relate to the general
nonlinear dynamics of the nancial system. Put simply, a nonlinear system is one in which a given change
in one variable may lead either to a small or a large change in another variable, depending on the level
of the rst variable. Nonlinear systems can sometimes exhibit very large changes in behaviour as a result
of very small alterations of key parameters or variables. And in some cases they are complex systems
for which it is impossible to predict long-term behaviour because an observer can never know the
relevant key values with sufcient accuracy. Furthermore, because economic systems are composed
of agents (such as individuals, rms and regulators) that interact in various ways, the dynamics of the
system can depend on precisely how these agents interact
11
. The scientic literature on the complex,
nonlinear dynamics of networked systems is currently in its infancy in terms of concrete predictions and
reliably generalisable statements. But, integrating existing knowledge with the foundations of nancial
economics literature can offer glimpses of important insights. Figures 4.2 to 4.6 illustrate unusual price,
quote and volume dynamics generated by algorithms consistent with feedback loops, both of the
amplifying (positive feedback loop) type and of the pinning (negative feedback loop) type.
7 See DR1, DR 12 (Annex D refers), Brogaard, (2010), Chaboud et al. (2009), and Hasbrouck & Saar (2011).
8 It is well-known that volatility alone accurately represents neither risk nor welfare (see, for example, Rothschild & Stiglitz
(1970) and Grossman, (1976)). Therefore, without further studies, no automatic and unambiguous welfare conclusions can be
reached from observations on volatility.
9 Jovanovic & Menkveld (2011), have a comparison of the volatility of Dutch and Belgian stocks before and after the entry of one
HFT rm in the Dutch market and nd that the relative volatility of Dutch stocks declines slightly.
10 For example, Zhang (2010), proxies for HFT with a measure of daily trading volume not associated with changes in quarterly
institutional holdings. Zhang nds an association between this trading volume measure and ‘excess’ volatility. While the proxy
likely relates to CBT, the correlation is difcult to interpret as arising from HFT because the volume-volatility relation appears
well before the adoption of HFT as currently dened. A stronger relation between volume and volatility can result from
increased welfare-enhancing investor risk-sharing. Therefore, indirect studies of HFT and CBT as in Zhang (2010) are not
recommended as a basis for formulating policy options.
11 See DR6, DR7 (Annex D refers), and Haldane & May (2011).
65
Legend
Tr ades
0
<10:40 10:41 10:44 10:45 10:46 10:47 10:48 10:49 10:50 10:52 10:54
1000 2000 3000 4000 5000 6000 7000 8000
2.75
3.75
4.75
5.75
6.75
7.75
8.75
The Future of Computer Trading in Financial Markets
66
Figure 4.2: Violent cycling of a stock with ticker CNTY in different USA exchanges on 21 June 2011
Legend
Tr ades
Best Bid
Best Ask
BDS
0
<10:40 10:41 10:44 10:45 10:46 10:47 10:48 10:49 10:50 10:52 10:54
1000 2000 3000 4000 5000 6000 7000 8000
2.75
3.75
4.75
5.75
6.75
7.75
8.75
ASKS
NSDQ
BOST
PACF
BATS
CNC
CBOE
EDGX
EDGE
BATY
NSDQ
BOST
PACF
BATS
CNC
CBOE
EDGX
EDGE
BATY
Quote and trade counts and Time (mm:ss)
P r i ce ( U S $)
Source: Nanex
Legend
Tr ades
Best Bid
Best Ask
BDS
0
<10:40 10:41 10:44 10:45 10:46 10:47 10:48 10:49 10:50 10:52 10:54
1000 2000 3000 4000 5000 6000 7000 8000
2.75
3.75
4.75
5.75
6.75
7.75
8.75
ASKS
NSDQ
BOST
PACF
BATS
CNC
CBOE
EDGX
EDGE
BATY
NSDQ
BOST
PACF
BATS
CNC
CBOE
EDGX
EDGE
BATY
Financial stability and computer-based trading
Figure 4.2 (cont.): Violent cycling of a stock with ticker CNTY in different USA exchanges on 21 June 2011
Legend
Tr ades
0
<10:40 10:41 10:44 10:45 10:46 10:47 10:48 10:49 10:50 10:52 10:54
1000 2000 3000 4000 5000 6000 7000 8000
2.75
3.75
4.75
5.75
6.75
7.75
8.75
Trade count and Time (mm:ss)
P r ice ( U S $)
Source: Nanex
67
The Future of Computer Trading in Financial Markets
68
Figure 4.3: Violent cycling of natural gas futures prices on NYMEX, 8 June 2011 with a resulting crash and subsequent recovery
Source: Nanex
Quote and trade counts and Time (hh:mm:ss)
P r i ce ( U S $)S i ze (n u m b e r of co nt r a c t s)
Financial stability and computer-based trading
Figure 4.4: An unexplained slump in European index futures on 27 December 2010
09:03:02
09:04:02
09:05:02
09:06:02
09:07:02
09:08:02
09:09:02
09:10:02
09:11:02
09:12:02
09:13:02
09:14:02
DAX Future
Time (hh:mm:ss)
EuroStoxx50 Future
-2.0
-1.5
Percentage change
-1.0
-0.5
0
Source: Deutsche Börse Group: Eurex Exchange
69
The Future of Computer Trading in Financial Markets
Figure 4.5: A pinning example that occurred during the recent Knight Capital algorithm malfunctioning episode where algorithms bought and sold
in a way that kept prices of a stock with ticker CODE glued (pinned) around $10.05
P r i ce ( U S $)
P r i ce ( U S $)
Vo lu m e ( Tr a d es)
Vo lu m e ( Tr a d es)
Time (hh:mm)
Time (hh:mm)
Source: Nanex
70
Legend
Tr ades
0
10:45 10:07:2510:07:1710:07:1510:07:1410:07:1310:07:10 10:07:10 10:07:42
250 500 750 1000 1250
1.00
2.17
3.33
4.50
5.67
6.83
8.00
Financial stability and computer-based trading
Figure 4.6: A crash in the stock AMBO that occurred and resolved itself within a single second
Legend
Tr ades
Best Bid
Best Ask
BDS
0
10:45 10:07:2510:07:1710:07:1510:07:1410:07:1310:07:10 10:07:10 10:07:42
250 500 750 1000 1250
1.00
2.17
3.33
4.50
5.67
6.83
8.00
ASKS
NSDQ
BOST
PACF
BATS
CNC
CBOE
EDGX
EDGE
BATY
NSDQ
BOST
PACF
BATS
CNC
CBOE
EDGX
Pr i ce ( U S $)
Trade count and Time (hh:mm:ss)
Source: Nanex
71
Legend
Tr ades
Best Bid
Best Ask
BDS
0
10:45 10:07:2510:07:1710:07:1510:07:1410:07:1310:07:10 10:07:10 10:07:42
250 500 750 1000 1250
1.00
2.17
3.33
4.50
5.67
6.83
8.00
ASKS
NSDQ
BOST
PACF
BATS
CNC
CBOE
EDGX
EDGE
BATY
NSDQ
BOST
PACF
BATS
CNC
CBOE
EDGX
The Future of Computer Trading in Financial Markets
72
Figure 4.6 (cont.): A crash in the stock AMBO that occurred and resolved itself within a single second
Legend
Tr ades
0
10:45 10:07:2510:07:1710:07:1510:07:1410:07:1310:07:10 10:07:10 10:07:42
250 500 750 1000 1250
1.00
2.17
3.33
4.50
5.67
6.83
8.00
P r i ce ( U S $)
Trade count and Time (hh:mm:ss)
Source: Nanex
Financial stability and computer-based trading
Figure 4.7: Hedge feedback loop
Index falls
Adjust
delta-hedge
Sale of stocks in
delta-hedge holding
While market crashes have always been a feature of nancial markets, the problem of understanding
the mechanisms behind system-level events in CBT environments is more recent. A good illustration
for the sort of systemic non-linear events that mechanical rule-following is able to generate can be
found in the portfolio-insurance-led market decline of 1987
12
. In order to hedge their risks, when
stock indices dropped, portfolio insurers were required to adjust their ‘delta-hedge’ holdings of stocks
used to balance risk. However, the values of the stocks in the delta-hedge holdings were used to
calculate the value of the index. So, when the index dropped, stocks held in the delta-edge portfolio
were sold, which depressed prices and pushed the index even lower; this then caused further sales
of the delta-hedge holdings, pushing the index still lower. This positive feedback loop (the effects of a
small change looping back on itself and triggering a bigger change, which again loops back and so on)
had a profoundly damaging effect, leading to major share sell-offs. This loop is illustrated in Figure 4.7
where a fall in the value of an index forces delta-hedgers to sell into a falling market, which in turn puts
downward pressure on the value of the index. Such destructive feedback loops can generate nonlinear
dynamics and can operate until delta-hedges no longer need to be adjusted or until a market halt is
called. The mechanical and unthinking execution of such ‘programme trades’ led in 1987 to huge selling
pressure and to price falls which were much deeper than were warranted by actual market conditions.
Forced sales in a nonlinear, self-fullling and sometimes self-exciting frenzy create a discrepancy
b
etween market prices and the true values of securities (measured by the fundamental value of the
asset) and thus constitute major market instability, which comes at a substantial social cost. It might be
argued that the more trading decisions are taken by ‘robot’ CBT systems, the higher the risk is of such
wild feedback loops in a nancial system. This endogenous risk is the logical thread that runs through
much of the rest of this chapter. The endogenous risk from human-programmed algorithms may differ
in important ways from feedback loops and risks in markets with greater direct human involvement.
4.3 How is computer-based trading thought to affect financial stability?
4.3.1 Mechanisms of instability
It seems unlikely that the future of CBT in the nancial markets leads merely to a faster system, and
therefore to more frequent crashes and crises, purely on the (metaphorical) basis that the ‘same
old movie’ is now being played at fast-forward speed. Rather, it seems more likely that, despite all
its benets, CBT has the potential to lead to a qualitatively different and more obviously nonlinear
nancial system in which crises and critical events are more likely to occur in the rst place, even in the
absence of larger or more frequent external fundamental shocks. Some of the insights into what the
precise mechanisms could be are outlined below.
Three key inter-related mechanisms that can lead to instability and losses can be summarised as follows:
• Sensitivity: systemic instability can occur if a CBT environment becomes more sensitive to small
changes or perturbations. If nancial dynamics in a CBT world become sufciently nonlinear so that
widely different outcomes can result from only small changes to one or more current variables (the
so-called ‘buttery effect’ in chaotic dynamics), then the observed prices and quantities are prone
to cascades, contagions, instability and path-dependency
13
. The mechanisms of deviation may be the
information sensitivity mechanisms, and/or the internal feedback loops, both of which are discussed
12 See DR9 (Annex D refers), and also Gennotte & Leland (1990).
13 See DR7 (Annex D refers), and also Haldane & May (2011), op. cit.
73
The Future of Computer Trading in Financial Markets
further below. Even if the effects were temporary and the original driving variables were to revert
to their long-term average values over time, some further irreversible events may have occurred,
such as nancial losses or even bankruptcies due to forced sales or to the triggering of penalty
clauses in contracts. At very high speeds, sensitivity may also be positively related to speed.
• Information: the existence of excessive nonlinear sensitivities can be due to informational issues: in
other words, who knows what and when. The information structure of a market has the potential
to increase or reduce market swings through a number of subtle, and sometimes contradictory,
effects. To illustrate this, academic studies have explored behaviour that arises in coordination games
with diffuse information
14
. In these game scenarios, agents coordinate to create a ‘bank-run’ on an
institution, a security or a currency if a given publicly observed signal is bad enough. Only very small
differences in the signal, say the number of write-offs of a bank, determine whether creditors run
or stay. With diffuse information, violent cascades of failures over an entire market system can be
triggered by such small events
15
.
• Endogenous risk: this term, commonplace in the nancial literature
16
, identies features wholly within
the nancial markets that lead in some situations to the sudden (due to the nonlinearities involved)
emergence of positive (i.e. mutually reinforcing) and pernicious feedback loops, whether market
participants are fully rational or otherwise
17
.
In the section that follows, a number of feedback loops that can contribute to endogenous risk are
explored. One of the reasons the sensitivities and feedback loops at very high speeds may look
qualitatively different from those at lower speeds, even if accelerated, lies in the fact that beyond
the limits of human response times, any feedback loops must be generated solely by robots
18
. It is
plausible that the properties of securities return time series at different frequencies are very similar to
each other through scaling
19
, at least as long as the frequency is low enough so that micro structure
effects do not come into the picture. This similarity may no longer hold beyond a critical frequency
at which only machines can react to each other. One of the reasons may be the discontinuity in the
set of possible feedback loops as one goes to very small scales and excludes human interactions
20
.
While ‘mini-ash crashes’ occur regularly at very small scales, they are mostly self-healing in a matter
of milliseconds and aggregate to noise without attracting much attention
21
. This does not mean that a
mini-ash crash of the feedback loop variety (as opposed to an execution algorithm simply splitting a
large trade and continuously buying or selling) could not form a chain reaction, gather momentum and
be visible at human scales also, or indeed lead human traders to feed into the loop.
14 See, for example, Carlsson & Van Damme (1993) and Morris & Shin (2002).
15 Information, sensitivity and speed interact. For instance, with lower tick sizes the frequency of order book updating increases
dramatically. Given possible and realistic latencies for traders, there would be inevitable information loss: no trader would at
any moment in time be able to have an accurate idea of the current state of the order book, let alone an accurate depiction
of multiple order books across trading venues. Very fast markets make concurrent precise order book information virtually
impossible and lead to trading decisions taken under a veil of incomplete knowledge, not only about the future but also about
the present. Trading strategies need to second-guess the state, possibly leading to over-reactions (see Zigrand (2005), for
a formal model). In that sense, speed inuences both information and sensitivity according to a U shape: perhaps, as with a
bicycle, both too low a speed and too high a speed lead to inefciencies and instabilities.
16 See Danielsson & Shin (2003), Danielsson et al. (2011) and Shin (2010).
17 See DR2, DR6, DR7, and DR9 (Annex D refers).
18 See DR6, DR7 and DR27 (Annex D refers). DR27 suggests a transition may occur at around 1000 microseconds.
19 This is the so-called characteristic of self-similarity, roughly saying for instance that the distribution of hourly returns is very
similar to the distribution of minute by minute returns when minute by minute returns are multiplied by the appropriate
scaling factor of 60 to a power H. For example in the Gaussian case H=1/2. In that sense, self-similarity means that short-dated
returns really are just a sped-up version of lower frequency returns, with no qualitative break in their statistical behaviour.
20 At sub-human scales all other AT activities also operate without direct human judgement, so one execution algorithm can
lead to many price-adjustment ticks in the same direction even without any feedback loops, leading to mini-ash crashes if
the order book is not concurrently replenished. The vast majority of such events heal themselves through limit order book
replenishment, leaving the crashes invisible at lower frequencies. If, on top of an execution algorithm at work, feedback loops
form at sub-human scales, then the dynamics become nonlinear and the crash may well be visible at human scales, unless
preemptive circuit breakers successfully kicked in.
74
21 Also see DR29 (Annex D refers). Most of the evidence in that review is based on US data.
Financial stability and computer-based trading
Figure 4.8: Risk feedback loop
Synchronised
selling
of risk
Initial Capital Prices
losses hit, risk adversely
increases affected
Losses on
positions,
“Haircuts”
go up
A better understanding is needed of the safety envelope within which such feedback loops either do
not arise or can be interrupted with condence and at low cost (see, for example, the case study of
risk management in the nuclear energy industry in DR26
22
). Amongst the prudential devices mooted
to reduce or cut feedback loops are various circuit breakers and trading limits that ideally limit false
positives (breaks that prevent the market from adjusting to the new fundamental value following an
informative piece of news) by being able to judge whether market stress is due to fundamentals or
feedback loops. The role of circuit breakers is discussed in more depth in Chapter 6.
Risk feedback loop
Financial crises typically involve endogenous risk of the following sort (see Figure 4.8). Assume that
some nancial institutions are hit by a loss that forces them to lower the risk they hold on their books.
To reduce risk, they need to sell risky securities. Since many institutions hold similar securities, the sale
of those assets depresses prices further. When institutions are required to practice ‘mark-to-market
accounting’ (where the value of some holding of securities is based on the current market price of
those securities), the new lower valuations lead to a further impact on bank capital for all institutions
holding the relevant securities, and also to a further increment in general perceived risk. Those two
factors in turn force nancial institutions to shed yet more of their risks, which in turn depresses prices
even further, and so forth. A small initial fundamental shock can lead to disproportionate forced sales
and value-destruction because of the amplifying feedback loops hard-wired into the fabric of nancial
markets
23
. Versions of this loop apply to HFT market makers: given the tight position and risk limits
under which high frequency traders operate, losses and an increase in risk lead them to reduce their
inventories, thereby depressing prices, creating further losses and risk, closing the loop. This value-
destruction in turn can cause banks to stop performing their intermediation role with adverse spill over
effects on the real economy. Devising measures to limit this loop ex-ante or to cut the loop ex-post is
notoriously difcult, not least due to the fallacy of composition
24
, although attempts to further diversify
among market participants should be encouraged.
22 DR26 (Annex D refers).
23 For further details see Brunnermeier & Pedersen (2009). That such feedback loops can lead to dynamics exhibiting fat left tails
(i.e. an increased probability of extreme losses) as well as volatility clustering has been shown in both rational, forward-looking
models (Danielsson et al. (2011)) as well as in models with myopic agents (see Danielsson et al. (2004), for a model where the
friction generating the feedback loops is given by a value-at-risk constraint; and Thurner et al. (2012), where the friction is a
margin constraint). For empirical evidence on the effect of losses or redemptions see Joshua & Stafford (2007). For evidence
on how shocks can propagate across different nancial institutions see Khandani & Lo (2007), and Khandani & Lo (2011).
24 This fallacy is discussed in Samuelson, P. (1947), and goes back at least to John Stuart Mill (1846). It refers to arguments that
attribute a given property to the whole only because the individual elements have that property. In the present example, it is
tempting to say that because each nancial institution sheds risk and becomes safer, the nancial sector
as a whole must be safer. Quite the opposite is true.
75
The Future of Computer Trading in Financial Markets
Figure 4.9: Volume feedback loop
Prices
fall
Investor Investor
HFTs
decides algorithm
“press-
to sell sells on
the-
increased volume
parcel”
Volume
increases
Volume feedback loop
Whether the analysis in the ofcial US Commodities and Futures Trading Commission/Securities and
Exchange Commission (CFTC/SEC) report
25
of the Flash Crash events of 6 May 2010 turns out to be
accurate and complete or not (see DR4 and DR7
26
for further discussion), it does illustrate a potential
driver of risk. The CFTC/SEC report describes a possible scenario whereby some HFT algorithms may
directly create feedback effects via their tendency to hold small positions for short periods. Such a ‘hot-
potato’ or ‘pass-the-parcel’ dynamic occurred on 6 May 2010 when trading amongst high frequency
traders generated very large volumes while hardly changing the overall net position at all (see Figure
4.9 where a sale leads to a price drop and an increase in HFT inventories which HFT then quickly try
to reduce, leading to increased trading volumes, which in turn encourage the original algorithm to sell
more). Because nancial instruments were circulating rapidly within the system, the increase in volume
triggered other algorithms which had been instructed to sell more aggressively in higher volume
markets (presumably on the basis that higher volume means lower market impact), selling into the
falling market and closing the loop
27
. Circuit breakers and algorithm inspection may prevent some of
these loops developing.
25 US Commodity Futures Trading Commission and the US Securities and Exchange Commission (2010).
26 DR4, DR7 (Annex D refers).
27 Kirilenko et al. (2011) provides evidence that some market participants bought and sold from each other very rapidly with
small changes in position over very short horizons during the Flash Crash. While this is a case study of only one (index)
security over a few trading days, it highlights the importance of the need for a better understanding of certain CBT and HFT
strategies and their interactions with each other and other market participants. While not able to directly identify CBT or
HFT, Easley et al. provide evidence of the speed and magnitude of unusual trading behaviour in more securities during the
Flash Crash, see Easley et al. (2011a).
76
Financial stability and computer-based trading
Shallowness feedback loop
Closely related is the potential feedback loop described by Angel (1994)
28
and Zovko and Farmer
(2002)
29
illustrated in Figure 4.10. Assume an initial increase in volatility, perhaps due to news.
The distribution of bids and offers in the order book adjusts and becomes more dispersed
30
. With
everything else constant, incoming market orders (i.e., orders to buy or sell at the market’s current
best available price) are more able to move the market reference price (based on the most recent
transaction price). This increase in volatility in turn feeds back into yet more dispersed quotes, and
the loop is closed.
Figure 4.10: Shallowness feedback loop
Volatility up
Bids and asks
become more
Incoming market
orders move prices
more dispersed
Order book imbalance (OBI) feedback loop
Some HFT algorithms trade by relying on statistical tools to forecast prices based on the order
book imbalance, roughly dened as the difference between the overall numbers of submitted bids
and submitted asks. Assume that a trader wishes to lay off a large position and adds asks at the best
ask. The order book now becomes imbalanced towards selling pressure, and high frequency traders
operating an OBI strategy add their own offers and remove some of their own bids given their
prediction of a falling price (the market falls without a trade, conrming the forecast), creating further
OBI and closing the loop (see Figure 4.11). The feedback is interrupted once fundamental traders (i.e.
traders with investment decisions based on fundamental valuation), who tend to be slower traders,
step in to purchase, anchoring prices again to fundamentals.
Figure 4.11: Order book imbalance feedback loop
OBI algorithms
predict future
downward pressure
Fundamental trader
Order book becomes OBI algorithms position
desires to sell and
more imbalanced towards dominant side
submits asks
28 Angel (1994).
29 Zovko & Farmer (2002).
30 For an examination of these types of book changes prior to news about earnings, see Lee et al. (1993).
77
The Future of Computer Trading in Financial Markets
News feedback loop
Many automated HFT systems work primarily on numeric information from market data sources
concerning prices and volumes of market orders, but some HFT systems include a ‘news listener’
component that scans headlines for tags and acts upon them immediately by broadcasting the tag
to all other components of the HFT system (news analytics, the computerised analysis of text-based
news and online discussions to generate CBT systems, is discussed in greater detail in Chapter 2 and in
DR8
31
). High frequency traders buy or sell depending on where prices differ from their own perceived
fair value; if the transactions of HFT systems are reported in news feeds, and picked up on by other
HFT systems, those systems can be led to revise their price in a direction that encourages them (or
other high frequency traders) to make similar trades (see Figure 4.12).
Figure 4.12: News feedback loop
HFT sells
Newsfeed reports
sale, HFT news listener
HFT “microprice”
revised down below
true market price pick up the story
Delay feedback loop
Eric Hunsader of Nanex Corp. suggests
32
the potential of the following very simplied feedback loop
which may have operated during the Flash Crash. Consider a fragmented market suffering from
overall selling pressure of a group of high-capitalisation stocks (for example originating from the sales
of E-Mini futures), and assume that the NYSE quotes lag slightly (see Figure 4.13). Since the market is
falling, the delayed NYSE bids appear to be the most attractive to sellers, and all sales are routed to
NYSE, regardless of the fact that actual bids were lower. High frequency traders that follow algorithmic
momentum strategies short those stocks and, given the unusual circumstances in the market, may sell
inventories. A second feedback loop then reinforces the rst one: as delays creep in and grow, the
increased urry of activity arising from the previous feedback loop can cause further misalignments
in bid/ask time stamps, so that the delay feedback loop amplies the pricing feedback loop
33
. The
provision of prompt public information on delays, possibly during a circuit breaker-induced pause,
would reduce the occurrence and violence of such loops. Furthermore, some throttles, such as order-
to-trade ratios, on excessive quote submissions that are intended to slow trading venue systems may
prevent some of these delays in the rst place.
Figure 4.13: Quote-delay feedback loop
Sell orders
routed to
NYSE
Delay
worsened
NYSE shows
best bid prices
NYSE falls,
NYSE quotes
delayed
31 DR8 (Annex D refers).
32 See DR7 (Annex D refers), and also: Hunsader (2010).
33 Consistent with delay feedback loops having stronger effects in securities where trading is spread across more markets,
Madhavan (2011) nds that, across stocks, the impact of the Flash Crash is positively related to measures of fragmentation in
the previous month.
78
Financial stability and computer-based trading
Index feedback loop
The CFTC/SEC nal report on the Flash Crash argued that the extreme volatility of the individual
component securities spilled over into the ETF (exchange-traded fund) markets and led market makers to
pause their market making activities. In response, the illiquid and stub ETF prices for indices provided false
systematic factor signals upon which factor models price individual securities on a risk-adjusted fashion,
feeding back into the pricing of individual securities, and thereby closing the loop (see Figure 4.14)
34
.
Figure 4.14: Index feedback loop
ETF market
affected by
illiquidity
Volatility Single-stock HFTs reduce linkage
up volatility up, between markets for
factors mispriced ETFs and single stocks
Market making
of single
stocks affected
While this chapter focuses on the sources of nancial instability in modern CBT environments, CBT
by itself can, of course, also be benecial; for example, it can lead to stabilising dynamics. Consider
the index feedback loop. When idiosyncratic volatilities create distorted valuations across venues and
products in normal times, high frequency traders engage in protable latency arbitrage by buying
low and selling high. They thus align valuations across venues and products, which leads to welfare-
improving allocations and prevents the dislocations of prices and the resulting loss of condence in
markets which could otherwise have become a fertile breeding ground for destabilising loops
35
.
Notice that while the unconditional average volatility may be reduced by CBT due to these stabilising
dynamics in normal times (and therefore much of the time), the existence of nonlinear reinforcing
feedback loops that can create very large swings under exceptional circumstances prompts further
caution about the simplistic use of volatility as a criterion for judging the health and safety of markets.
4.3.2 Socio-technical factors: normalisation of deviance
Cliff and Northrop propose in DR4 that the Flash Crash event in the US nancial markets on 6 May
2010 is, in fact, an instance of what is known as ‘normalisation of deviance’ and they explain that such
failures have previously been identied in other complex engineered systems. They argue that major
systemic failures in the nancial markets, on a national or global scale, can be expected in future, unless
appropriate steps are taken.
Normal failures (a phrase coined in 1984 by Charles Perrow)
36
in engineered systems are major
system-level failures that become almost certain as the complexity and interconnectedness of the
system increases. Previous examples of normal failure include the accident that crippled the Apollo 13
moon mission, the nuclear-power accidents at Three Mile Island and Chernobyl, and the losses of the
two US space shuttles, Challenger and Columbia.
34 Madhavan (2011) op. cit. analyses the feedback loop between ETFs and the underlying securities during the Flash Crash.
35 One piece of evidence that points towards the role of high frequency traders as removers of (transitory) pricing errors can be
found in the working paper, Brogaard et al. (2012).
36 Perrow (1984).
79
The Future of Computer Trading in Financial Markets
As Cliff and Northrop note
37
, the American sociologist Diane Vaughan has produced detailed analyses
of the process that gave rise to the normal-failure losses of Challenger and Columbia space shuttles,
and has argued that the key factor was the natural human tendency to engage in a process that she
named
38
‘normalisation of deviance’: when some deviant event occurs that was previously thought to
be highly likely to lead to a disastrous failure. If it then happens that actually no disaster does occur,
there is a tendency to revise the agreed opinion on the danger posed by the deviant event, assuming
that, in fact, it is normal: the deviance becomes normalised. In essence, the fact that no disaster has yet
occurred is taken as evidence that no disaster is likely if the same circumstances occur again in future.
This line of reasoning is only broken when a disaster does occur, conrming the original assessment of
the threat posed by the deviant event.
Cliff and Northrop argue that the Flash Crash was, at least in part, a result of normalisation of
deviance. For many years, long before 6 May 2010, concerns about systemic effects of rapid increases
in the price volatility of various instruments had led several UK exchanges to implement circuit breaker
rules, requiring that trading in a security be suspended for a period of time if the price of that security
moved by more than a given percentage within a short time. In response to the Flash Crash, the US
SEC has now imposed similar mechanisms in the US markets with the aim of preventing such an event
re-occurring. Thus, it seems plausible to argue that before the Flash Crash occurred there had been
a signicant degree of normalisation of deviance: high-speed changes in the prices of equities had been
observed, market participants were well aware that they could lead to a high speed crash, but these
warning signals were ignored and the introduction of safety measures that could have prevented them
was resisted.
Moreover, it could plausibly be argued that normalisation of deviance has continued to take place
in the markets since the Flash Crash. There are anecdotal reports (summarised in DR4) that the
speed of price uctuations occurring within the limits of circuit breaker thresholds seems to be
increasing in some markets; and there is evidence to suggest that another ash crash was avoided
on 1 September 2010, in a similar period when quote volumes exceeded even those seen at peak
activity on 6 May 2010, but no ofcial investigation was commissioned to analyse the events of
September 2010. Furthermore, the circuit breaker mechanisms in each of the world’s major trading
hubs are not harmonised, exposing arbitrage opportunities for exploiting differences; computer and
telecommunications systems can still fail, or be sabotaged. The systemic effects of these failures may
not have been fully considered.
Of course, the next ash crash will not be exactly the same as the previous one. But there are no
guarantees that another event, just as unprecedented, just as severe, and just as fast (or faster) than the
Flash Crash cannot happen in future. Normalisation of deviance can be a very deep-running, pernicious
process. After Challenger, NASA addressed the immediate cause (failure of a seal on the booster
rockets), and believed the Shuttle to be safe. That was no help to the crew of Columbia. Reassurances
from regulators are likely to sound somewhat hollow since the next market failure may well be a failure
of risky technology that, like the Flash Crash, has no clear precedent.
Cliff and Northrop (DR4) argue that the normalisation of deviance poses a threat to stability in the
technology-enabled global nancial markets. Indeed, the dangers posed by normalisation of deviance
and normal failures are, if anything, heightened in the nancial markets because the globally
interconnected network of human and computer traders is what is known in the academic literature
as a socio-technical system-of-systems. In other words, it is an interconnected mesh of people and
adaptive computer systems interacting with one another, where the global system is composed of
constituent entities that are themselves entire systems, with no single overall management or
coordination. Such systems are so radically different from traditional engineered systems that there is
very little established science or engineering teaching that allows us to understand how to manage and
control such super-systems. This issue of normalisation of deviance in the nancial markets and its role
in informing the discussion of possible regulatory options was recently discussed in more detail by
80
37 DR4 (Annex D refers).
38 Vaughan (1997).
Financial stability and computer-based trading
Haldane (2011)
39
. A comparison between the systemic risk and safety in the nuclear industry and in
nance can be found in DR26.
4.4 How can computer-based trading affect financial stability in future?
4.4.1 Forces at play
The strength of feedback loops depends on a number of variables, especially the capitalisation levels
and the leverage ratios of nancial institutions, and the degree of diversity of market participants. For
instance, if liquidity is provided by lightly capitalised HFT operators rather than market makers with
large inventories
40
, then the passing-of-the-parcel scenarios as well as the resulting feedback loops are
stronger because inventory management with little capital requires the rapid ofoading of positions
when prices fall, creating a volume feedback fall in prices. In that sense at least, substituting speed (with
the cancellations that come with speed) for capital works well for market making purposes in normal
conditions but may not work well in times of greater stress where the lack of capital can contribute
extremely quickly towards positive instead of negative feedback loops
41, 42
.
Similarly, the race for speed can reduce market diversity if market participants adopt the same (minimal
number of) lines of code, leading to algorithmic crowding (see DR27). If similar mistakes creep
into those algorithms because of inherent programming or engineering difculties (DR26), the less
diversied market participants may act in unison, creating stronger feedback loops, especially if the
substitution of speed for capital further erodes investor condence.
Moreover, diversity itself can worsen during an episode of endogenous risk as an unintended
consequence of the combined interactions of risk-management systems, regulatory constraints, margin
calls and mark-to-market accounting requirements. These can lead to instantaneous synchronisation
of actions among a group of institutions if they are all subject to the same regulations, constraints, and
coordination mechanisms. For example, the CFTC and SEC found that during the crucial minutes of the
Flash Crash, liquidity providers switched to becoming liquidity demanders and sold aggressively and in
an unsophisticated fashion into a falling market once their inventories reached a certain level. Where
liquidity provision is in the hands of a small number of large but lightly capitalised market participants,
this can lead to a rapid drying-up of liquidity (see also DR7 for an estimate of the strength of this
effect). In fact, informational asymmetries (where one participant knows that they know less than
another) have the power to strengthen the endogenous risk mechanism in a variety of ways (for more
details, see DR9
43
). This is because investors may mistake a temporary forced sale for a sale based on
negative privileged information, leading to even lower prices.
Furthermore, information in CBT environments can present many subtle dangers. For example,
even if all active market participants know that a certain event has not occurred, market prices
and quantities would still not be able to completely discount the event, for even if it is known by all
that the event has not occurred, there can still be participants who do not know that others know
that it has not occurred. It can be argued that technology has further removed, to a certain extent,
common knowledge from markets. This is also a recurring theme in the interviews of computer-based
traders conducted by Beunza et al. (IN1)
44
. Markets have become networked, distributed computing
environments (and a well-known theorem states that events cannot be common knowledge in
distributed computer environments due to the absence of concurrent centrality of observation, a
phenomenon sometimes referred to as the ‘electronic mail game’)
45
. This has, in effect, introduced
further levels of complexity, if only because the outcome of the market now depends in a non-trivial
39 Haldane (2011).
40 For empirical evidence on the importance of liquidity providers’ balance sheets see Comerton-Forde et al. (2010).
41 For further details on how intermediaries manage risk and how this affects price dynamics see Dufe (2010). For empirical
measures of the magnitude see Hendershott & Menkveld (2011).
42 It seems that, in terms of market structure, the presence of low latency ‘cream skimming’ liquidity providers with little capital
has driven out market makers with capital and inventories. Unfortunately, there is little empirical work available on how
pervasive ‘cream skimming’ really is; see Menkveld (2012).
43 DR9 (Annex D refers).
44 IN1 (Annex D refers).
45 See: Rubinstein (1989), and Halpern & Moses (1990).
81
The Future of Computer Trading in Financial Markets
way on what each trader believes any other trader believes about all other traders’ beliefs, and so on.
The direct link between market outcomes and the fundamental events that ought to act as anchors
for valuation has been further severed and replaced by a complex web of iterated and nested beliefs
46
.
Meanwhile, there may have been a weakening of shared informal norms (such as the implicit sense of
obligation to refrain from certain actions or to step in and help out). Thus, each actor has a very limited
view of the entire landscape and limited interaction with other humans
47
.
Interviews conducted by Beunza et al. (IN1) suggest that, as a result of these shifts, computer-based
traders have learned to use social networks both in case of failure in the electronic networks and as a
means to access information not available ‘on the wire’, relating to for example, as market sentiment or,
more importantly, participation. According to Beunza et al. (IN1), they rely on such networks to form
views on who is still trading, what are volumes, are people taking risks, are prop desks busy, or are they
sitting on their hands waiting, trying to get a feel for markets
48
. The information gathered from such
links in turn informs the trader’s decision on whether to kill the algorithms.
This is not to say that full transparency is always ex-ante optimal. Ex-ante transparency can have
signicant costs in general, since risk-sharing opportunities are prevented, a phenomenon commonly
referred to as the Hirshleifer effect
49
. In pre-trade, for example, transparency may increase market
impact costs (also called implementation shortfall; for example, due to order anticipation or due to
informational considerations) and therefore lead ex-ante to less demand for securities that are more
difcult to trade, in turn making capital-raising by various entities and corporations more expensive,
leading to less liquid markets, and so on. Given the current market structure, dark markets (markets
without pre-trade transparency) have a role to play in normal times. In periods of stress, however, it
may well be that the social costs arising from the absence of common knowledge are large enough to
warrant an institutional mechanism designed to remove lack of transparency quickly in such periods
and illuminate all markets, as suggested in DR9.
To summarise, computer-based markets may introduce an additional obfuscatory layer between events
and decisions. This layer harbours the risk that unfounded beliefs or rumours may contribute to self-
reinforcing cascades of similar reactions and to the endogenous risk, pushing valuations away from
fundamentals.
4.4.2 Risks for the future
Left to its own devices, the extent of CBT may still grow, and the endogenous risk factors described
above will continue to apply. Some aspects of CBT, however, may have started to nd their level.
For instance, there are natural bounds on the extent of trading that can be generated by proprietary
(‘short-termist’) high frequency traders. First, those trades that have such traders on both sides of
a transaction can generate prots only for one. Second, the no-trade theorem
50
would predict that
trade will collapse once it becomes known that the only trades are those offered by short-termist
proprietary traders who do not have an incentive to hold the securities for fundamental reasons. And
third, much of the prots and rents of HFT are competed away under increasing competition. Recent
reports suggest that prots of HFT companies have declined
51
, and a study
52
in 2010 has established
that the total prots available for extraction by HFT may not be as large as some people suspect.
Looking at trading patterns, there is preliminary evidence that HFT may have reached its equilibrium
penetration in London and EuroNext equity trading (see DR5). Practitioner views seem to conrm this
46 For a detailed analysis, see Shin (1996).
47 This partial view has the potential in turn to lead, in turn, to actions based on misinterpretations and to unintended
consequences, since no actor can possibly have forecast, considered and internalised all the direct and indirect effects of their
actions, including all reactions and reactions to reactions and so on, see DR11 (Annex D refers). The ‘culture of restraint’
amongst NYSE specialists that led members to follow the norms even if it was against their self-interest, as documented by
Abolaa, (2001), would not be sustainable in a CBT environment where social norms become weaker and deviancy can set in,
as in the ‘broken window’ theory. For example, in the case of quote-stufng: if a oor trader had created deliberate noise by
rescinding hand signals in previous decades he would probably have been rapidly stopped from doing so.
48 IN1, p. 18 (Annex D refers).
49 Hirshleifer (1971).
50 Milgrom & Stokey (1982).
51 See, for example, Cave (2010).
52 Kearns et al. (2010).
82
Financial stability and computer-based trading
scenario
53
. Thus, CBT may gain market-share as more buy-side investors use it, but HFT, in the sense
of proprietary intermediation trading, is naturally limited by the fundamental trading volume of real-
money investors.
There are a few factors, however, that could further increase HFT. First, if proposals to remove some
of the dark pool waivers are adopted, institutional investors may be forced to trade more on lit pools.
Second, some public markets still retain an element of open-outcry, which may be eroded. Third,
the progressive migration of some over-the-counter (OTC) markets towards clearing may ultimately
lead to standardisation and trading on exchanges. Fourth, if market making activities by banks were
to fall under proprietary trading, more of the market making would be done by unafliated and lightly
capitalised high frequency traders. Finally, HFT will continue to expand into new public markets,
including emerging markets and electronic retail market places such as today’s Amazon or eBay
54
.
To conclude, over the next ve to ten years, there is likely to be further substitution of open-outcry,
dark and OTC trading in favour of CBT, rendering all the effects described above more salient still.
One aspect apparent in the interviews by Beunza et al. (IN1) is the feeling of ‘de-skilling’ among market
participants, especially on the execution side. As trading is done increasingly by algorithms, traders
are being replaced by ‘anti-traders’ whose role it is to shut trading engines down as well as to tweak
them, rather than to take decisions: “Traders gradually lose the instincts and tacit knowledge developed
in oor-based trading”
55
. With fewer traders actively involved, and also with fewer observers of the
markets due to automation, the chances of identifying risks early may be reduced.
Financial institutions optimise their trading subject to the regulatory environment, which means
that constraints will often be binding and may inuence market dynamics in unexpected and often
detrimental ways. For example, the academic literature has identied many circumstances where the
‘fallacy of composition’ appears: the market system is unstable despite the fact that each algorithm in
isolation is stable. This strongly suggests that if a testing facility for algorithms were to be introduced,
the safety of individual algorithms is not a sufcient or indeed even a necessary criterion for systemic
stability. It follows that, when thinking about the future of CBT and nancial stability, assumptions have
to be made as to the future regulatory and other constraints imposed upon markets and the new
dynamics carefully assessed. For example, further in-depth studies may provide indications as to how
minimum resting times or minimum tick sizes affect nonlinear market dynamics and nancial stability.
A second institutional feature which will affect future stability is the market segmentation between
the various competing trading venues. Aligning valuations for single securities, multiple securities, and
derivatives across venues is a socially useful task that high frequency traders currently perform for a fee.
Social welfare requires that this role be lled in a permanent and well-capitalised fashion. There is a
small risk that HFT will not be able or willing to perform this task in periods of market stress, margin
calls or collateral pressure.
Non-linearities in liquidity provision (leading to quick reversals between feast and famine) are an
important cause of system-wide non-linear dynamics that deserve further study. Most of the time
HFT adds to liquidity, but some of the time (in periods of stress or crisis) it subtracts liquidity, causing
price discontinuities. These liquidity non-linearities have probably become more acute in an HFT
world because of the effects discussed above, as well as the fact that algorithms are ne-tuned to
the time-series on which they are tested and will be taken off-line when faced with sharp price
movements, lack of liquidity or delays that render the models inapplicable. These considerations make
the ‘market makers’ problem’ of inventory and information management not only different but also
altogether more difcult.
Finally, in a world with multiple trading and pricing venues that are interconnected by high frequency
traders, the network topology determines the stability and the ow of information and trades.
Together with company-owned dark pools (broker crossing networks), the aggregate liquidity across
all venues may well be larger than with single monopolised exchanges, but the dynamic behaviour
of liquidity will depend more and more on the network structure as well as on the specics of the
53 WR3 (Annex D refers).
54 See Jopson (2012).
55 IN1, p. 18 (Annex D refers).
83
The Future of Computer Trading in Financial Markets
HFT operators which link the trading venues. A liquidity shock on one venue, that might have gone
unnoticed if there was one large centralised exchange, can now affect prices on that venue. In normal
times, the aberrant price would quickly disappear as cross-trading-venue high frequency traders buy
low and sell high. But in stressed markets, their capital may be limited, or the traders themselves may
start to doubt the prices (as happened during the Flash Crash) and refrain from arbitrage. Institutional
investors then start to mistrust valuations across the board, and the resulting pressures mean that high
frequency traders no longer contribute to liquidity provision, which makes price divergence across
trading venues worse still. And so the shock is transmitted through the network, and its effects are
reinforced by yet another positive feedback, as illustrated in Figure 4.15. Trades and transactions will
occur at socially inefcient prices, and mark-to-market valuations can only be done to multiple and
illiquid marks.
Figure 4.15: Systemic divergence feedback loop
Prices diverge
across multiple
market venues
Initial liquidity
HFTs reduce Market stress,
shock on one
linkage among illiquidity,
trading venue, HFT
trading venues real money disappears
capital strained
HFTs take
a hit
Understanding how to avoid such situations, and to contain them when they do occur, is a topic not
only for further research but for urgent consideration by policy makers and regulators. As argued in
Chapter 6, the reality and risks of multiple trading venues suggests that authorities, working together
with industry and academics, should develop proposals for a system of cross-market circuit breakers.
The degree of integration is open for debate. A loose collection of circuit breakers or limit up/limit
down restrictions may in fact create greater endogenous risk. For example, when market prices on
one security in one market are stuck on a limit down or limit up boundary while related prices on
other trading venues are not, linkages are broken and price coherence made impossible, leading to the
possible activation of the systemic divergence feedback loop. On the other hand, a break on a minor
market triggering a break on a major venue can create unnecessary confusion and uncertainty.
Furthermore, interviewees in the study by Beunza et al. (IN1) stated that CBT actions during a
dislocation to a primary market in a benchmarked price system will in part be determined by the
outcomes of social communications during the event. Depersonalisation also may lead to weaker social
norms and to less common knowledge that in turn can give rise to deviancy
56
and reduce investor (as
well as market maker) condence. The lack of condence may lead to shallower markets, which in
turn create less common knowledge and more opportunities for manipulation (see DR20
57
for further
details on market abuse), see Figure 4.16.
84
56 For example, various presumed low latency market manipulation strategies that operate across markets, and which are
therefore harder to pinpoint, have attracted attention. One possible such strategy (dubbed ‘The Disruptor’ by Nanex) would
work as follows. It relies on the fact that arbitrage algorithms keep the E-Mini futures and the ETFs on the S&P500 in line. The
strategy disrupts the E-Mini futures during a time where the ETF market is soft (or is made soft by the algorithm) by selling
excessive E-Minis and expecting the other side to have been buying, and therefore it can forecast that the other side will
hedge in the ETF market. The speed advantage of the originating algorithm means that it may purchase from the arbitrageurs
at a prot-making price.
57 DR20 (Annex D refers).
Financial stability and computer-based trading
Figure 4.16: Systemic market structure feedback loop
Less common
knowledge
Fragmented markets
HFT intermediation
More scope for
Markets more
manipulation,
substituting speed
shallow
for capital
weaker social norms
Loss of
confidence
ShaIlower markets may increase the possibility of various feedback mechanisms and therefore systemic
risk, and vice versa
58
. It would seem that policy-makers ought to take a step back from incremental
regulatory adjustment and rethink the deeper fundamentals of healthy markets. For example, various
proposed amendments to the market structure, such as Basel III and the bank ring-fencing proposals in
the Vickers Report, may imply that broker-dealers are likely to reduce capital and funding liquidity allocated
to market making, thereby leaving markets shallower and more vulnerable to endogenous risk
59
.
4.5 Conclusions
Markets with signicant proportions of computer-based high frequency traders are a recent
phenomenon. One of the most novel aspects of their dynamics is that interactions take place at a
pace beyond the capacity of human intervention, and in that sense an important speed-limit has been
breached. Because of the speed advantages that computers can offer CBT is now almost obligatory.
This gives rise to the potential for new system-wide phenomena and uncertainties. One key issue
is that information asymmetries become more acute (and indeed different in nature)
60
than in the
past; and the primary source of liquidity provision has changed, to computer-based and HFT systems,
which has implications for the robustness of markets in times of stress.
Research thus far provides no direct evidence that HFT has increased volatility. But, in certain specic
circumstances, self-reinforcing feedback loops within well-intentioned management and control
processes can amplify internal risks and lead to undesired interactions and outcomes. These feedback
loops can involve risk-management systems, and can be driven by changes in market volume or
volatility, by market news and by delays in distributing reference data.
A second cause of market instability is social: normalisation of deviance, a process recognised as a major
threat in the engineering of safety-critical systems, such as aeroplanes and spacecraft, can also affect the
engineering of CBT systems.
The next chapter discusses market abuse in the context of CBT.
58 The evidence on the question of whether HFT leads to markets that are less deep is not fully conclusive. DR5 tentatively
attributes to HFT part of the signicant increases in FTSE 100 mean and median LSE order book depths displayed at the best
quotes between January 2009 and April 2011, although the authors are silent on depth beyond the best quotes. They also
show that FTSE small cap depths have improved much less. Using different realised impact measures, DR1 nds little change in
realised impact over the same period. Using more recent data, Nanex nds huge variation in the depth of the rst ten layers
of the E-Mini futures contract with a possible downward path over the period from April 2011 to April 2012.
59 Shallowness operates directly through the shallowness feedback loop and the systemic market structure feedback loop, but
in general it accelerates all loops that rely on price adjustments, such as the hedge feedback loop, the risk feedback loop, the
volume feedback loop and so on.
60 The paper by Biais et al. (2011), focuses on asymmetric information arising not from the fact that some traders uncover new
information but from the fact that their lower latency simply allows them to have foreknowledge about information that will
be hitting markets a fraction of a second later anyway.
85
5 Market abuse and
computer-based trading
Key findings
Economic research thus far provides no direct evidence
that high frequency computer-based trading has increased
market abuse. However, most of this research does not
focus on the measurement of market abuse during the
continuous phase of trading.
Claims of market manipulation using high frequency trading
techniques are consistently reported by institutional
investors (pension funds or mutual funds) internationally.
The same investors describe having little or no condence
in the ability of regulators to curb the behaviour they
describe as abusive.
Policy makers should take such perceptions seriously,
whether or not they are at variance with reality: the true
extent of abuse will never be known and it is perceptions
that determine trading behaviour and investment decisions.
Increasing the ability of regulators to detect abuse and
producing statistical evidence on its extent would help to
restore condence.
87
The Future of Computer Trading in Financial Markets
5 Market abuse and
computer-based trading
5.1 Introduction
The London Interbank Offered Rate (LIBOR) scandal supplies an excellent example of attempted price
manipulation on a seemingly very large scale. It reminds us that forms of manipulation not conducted
using electronic means are very much alive. However, a possible link between high frequency trading
(HFT) and market abuse is frequently alluded to in the nancial press, in regulatory reports and in
direct interviews with investors. This link is the focus of this chapter: do increased latency differentials
between agents increase the likelihood of market abuse and market manipulation in particular? Taking
market abuse broadly to comprise insider trading and market manipulation, this chapter discusses
several, not mutually exclusive ways in which HFT may matter to market abuse.
There is currently very little large-scale and data-based evidence on the incidence of abuse, whether
in pre-HFT markets or today. However, qualitative evidence is accumulating that institutional investors
perceive abuse as more widespread in today’s markets, where HFTs may dominate liquidity supply.
This perception has in itself the potential to affect the behaviour of slower and less-informed agents,
particularly liquidity suppliers, with a consequent negative impact on market outcomes.
The concern that high frequency traders use a speed advantage over other agents to implement new
abusive strategies generates much speculation, but other issues may be equally important. In particular,
the growth of HFT may have altered the trading environment in ways that render some forms of abuse
easier to perpetrate than in the past.
As a result, two courses of action seem available:
• Perception should be conrmed or corrected through the production of statistical evidence
on the link between HFT and abuse.
• Signicant investment should be made by regulators to acquire the skills and create the
infrastructure that would allow them to deal with the surveillance issues that HFT has created
and thus help to build the condence of institutional investors.
5.2 Past: what was the impact of computer-based trading on market abuse
in recent years?
This section does not review large-scale and data-based evidence on the past incidence of abuse as
such evidence simply is not available to academic researchers: abuse is rarely identied unambiguously
and only a few cases every year are successfully prosecuted, even in the largest markets. For example,
Figure 5.1 shows that the total value of nes for market abuse that have been levied on nancial rms
by the UK regulator since 2003 show year-to-year variability of the total ne and a general lack of
trend. Where the total ne was large, it was dominated by a single massive ne
1
. No nes were levied
in 2008
2
.
88
1 Except in 2012 where three nes are clear outliers and make up the purple area of the bar.
2 It is important to note that because of the delays involved in detection and prosecution, the year in which the ne was
imposed was typically not the year during which the related abuse was conducted.
Market abuse and computer-based trading
Figure 5.1: Fines levied by the UK Financial Services Authority (FSA) for market
abuse, 2003-2012
0
2
6
10
14
4
8
12
16
18
20
Fines (£m)
2003/04 2004/05 2005/06 2006/07 2007/08 2008/09 2009/10 2010/11 2011/12
Year
Fines for market abuse excluding top fine
Top fine
Source: FSA Annual Reports
None of the nes imposed by the UK Financial Standards Authority (FSA) so far appears to have been
directly related to high frequency strategies, even though HFT has grown in importance over the past
ve years
3
.
The USA has seen a very small number of cases of market abuse related to HFT. In late 2010, the US
Financial Industry Regulatory Authority (FINRA), an industry body that has regulatory authority over
its member rms, announced that it was ning Trillium Brokerage Services $1m, for order book
layering. The relative lack of HFT-related cases is consistent with the interpretation that HFT is not
giving rise to more abuse, or alternatively that such abuse is much harder to detect. It is certainly
the case that the few penalties related to HFT pale into insignicance when compared to the nes
imposed to date on the rms implicated in the low frequency LIBOR xing scandal uncovered in 2012.
In May 2011, the FSA announced its intention to impose an £8m ne on Swift Trade, a now dissolved Canadian rm catering
for day traders, for order book layering, but the outcome of this decision is still pending.
89
3
The Future of Computer Trading in Financial Markets
They included the largest single nes ever levied by the FSA and US Commodity Futures Trading
Commission (CFTC). Barclays was ned £59.5m by the FSA, $200m by the CFTC and a further $160m
by the US Department of Justice.
Other data also suggest that the nes for market abuse are relatively insignicant. Table 5.1 shows the
number of criminal sanctions for market abuse imposed in some European Union (EU) Member States
between 2006 and 2008
4
. The numbers are clearly low and indicate how unlikely successful criminal
prosecution has been.
Table 5.1: Annual number of criminal sanctions for market abuse in various EU member
states, 2006-2008
Member State 2006 2007 2008
Austria N/A N/A 21
Belgium 1 2 1
Cyprus N/A N/A 6
France 31 19 16
Germany 20 7 16
Italy N/A 11 5
Luxemburg 0 0 0
Netherlands 1 2 4
Poland 4 11 8
Spain N/A 14 11
UK 6 1 6
There is a widely held suspicion that the number of these cases is vastly underestimated. Although
there is a lack of signicant empirical evidence, regulators and economists have given much thought to
the denition and implications of market abuse. These views are considered in the next section.
5.2.1 Regulatory and economic views of abuse
Manipulation can be achieved by various means, such as appearing informed (for example, blufng), or
through trading that temporarily pushes prices away from equilibrium, or by entering quotes that give a
false impression of actual liquidity. Market abuse is distinct from securities fraud, which typically implies
improper conduct on the part of a nancial market professional towards their client. In cases of market
abuse, abusers rarely know their counterparties and, even more importantly, abuse operates via the
price process of securities.
In order to prosecute, regulators need to identify victims of abuse. The regulatory approach to abuse
denes victims as those who have been penalised either through an unfair informational advantage or
through articially manipulated prices. Economists, on the other hand, study abuse because to them it
has market-wide implications for liquidity, turnover, pricing efciency and other variables affecting
social welfare
5
.
Both economists and regulators agree that the perception of abuse is crucial. One obvious reason for
this is that the true incidence of abuse can never be established, and current and potential investors
must therefore form an estimate of the likelihood of losing out unfairly against other counterparties.
As a result, a high perceived likelihood of suffering a loss through interaction with other traders may
have damaging implications for liquidity, efciency and transparency. Abuse has obvious direct costs
for its immediate victims, who suffer reduced investment performance via increased trading costs. But
perceptions of abuse also affect the behaviour of liquidity suppliers who will protect themselves against
4 European Commission Staff Working Paper (2011a).
5 This is why economists are not concerned with securities fraud, which they consider a purely regulatory and legal matter.
90
Market abuse and computer-based trading
expected trading losses by quoting poorer prices – the equivalent of an insurance premium. As a result,
increased trading costs and poorer performance of investment and pension funds may spread to most
other participants.
Another concern is that perceived abuse can reduce the equity market participation of large and less-
informed investors who have a longer investment horizon. Perceived abuse may also have an impact on
the market structure by inuencing investors’ choice of modes of trade. It may give incentives to large
and long-term investors to shun centralised, transparent markets favour of ‘private’ execution of their
trades, harming market transparency and price formation
6
.
An illustration of this appears in the International Organisation of Securities Commissions (IOSCO) (2011):
Some market participants commented to regulators that a loss of condence in the
fairness of markets may result in them reducing their participation in lit markets, or in
shifting their trading into dark pools. If this effect is indeed occurring on a signicant scale,
it may result in a growing concentration of trading (at least in the lit markets) amongst a
relatively limited set of market participants
7
.
Clearly then, perceptions of abuse should not be dismissed.
5.3 Present: what is the current impact of computer-based trading
on market abuse?
The commissioned review DR28
8
documents the changes in order ow and market structure that have
taken place in the past decade and points to the possible risks that this may raise for market abuse and
the potential for perpetrators to hide their actions. Preliminary empirical evidence is presented in three
reviews commissioned for this report (DR22, DR24 and DR28)
9
. These studies use publicly available
data to construct proxies for HFT activity and abuse.
The effects of HFT on end-of-day price manipulation on a large sample of exchanges around the world
are tested directly in DR22 and the authors conclude that the data demonstrate that HFT leads to
lower incidence of manipulation. Another study, using data from the London Stock Exchange (LSE) and
Euronext Paris from the period 2006–2011 also reports a signicant negative relationship between
HFT activity and end-of-day price manipulation (DR24). A longer span of Euronext Paris and LSE data
are used in DR28 to verify that HFT improves market efciency and integrity. Overall, these studies
provide a useful starting point from which to argue that fears of HFT generating abuse are unfounded.
However, the studies focus, in the main, on a very specic part of the trading day (the market close)
and do not attempt to measure the extent of market abuse during the continuous phase of trading.
Other forms of evidence have recently begun to appear. This evidence can be described as qualitative
or ‘soft’ in that it does not allow the quantication of the reality of abuse as such but pertains entirely
to perceptions of it.
6 In this respect, institutional investors must not be pitted against retail investors. The former often represent the pension and
investment savings of households, and any trading costs suffered by large funds will ultimately be borne by private savers.
7 International Organization of Securities Commissions (2011), p. 29.
8 DR 28 (Annex D refers).
9 DR22, DR24 and DR28 (Annex D refers).
91
The Future of Computer Trading in Financial Markets
5.3.1 Evidence on the perceptions of abuse
Evidence on investor perceptions of abuse has been published in the specialised press. It can also be
extracted from reactions to regulatory consultations and calls for evidence. Even more useful, because
they are conducted on a larger scale, are several recently conducted surveys of institutional investors,
which either touch upon or explore the link between HFT and abuse
10
. This chapter draws on several
surveys in which large investors claim that the liquidity supplied by HFT is thin, unbalanced and often
even ‘illusory’
11
. Many respondents were critical of HFT, some vehemently so. For example from the
Kay Review: “The majority (…) doubted that the liquidity which high frequency trading claimed to
provide was real”
12
.
From this, a claim often follows that such liquidity is not just the result of the nature of HFT-dominated
order ow but in fact amounts to deliberate manipulation:
There is broad agreement among traditional investors that abusive trading practices need
to be tackled, with most concerns centring on the high number of computer-generated
orders being used to manipulate ‘real’ liquidity
13
.
A survey of large investors and HFT conducted for the Swedish Government similarly notes that:
concern for market abuse is considerable (…) The majority of companies that were
surveyed expressed concern that a large portion of high frequency trading was being used
to manipulate the market. There are clear apprehensions that market abuse has become
more extensive and difcult to identify as a result of the sharp increase in the number of
orders and trades
14
.
Similar conclusions were drawn in analysis of the responses to the consultation of the IOSCO, which
contains two questions directly relevant to abuse and HFT
15
:
• Q9: Do you think existing laws and rules on market abuse and disorderly trading cover computer-
generated orders and are relevant in today’s market environment?
• Q10: Are there any strategies employed by HFT rms that raise particular concerns? If so, how
would you recommend that regulators address them?
All the entities that responded “no” to Q9, indicating that they felt inadequately protected from new
forms of abuse inicted by HFT, were bodies representing the buy side. Similarly, the institutions
representing the buy side answered overwhelmingly “yes” to Q10 (followed by description of
specic concerns).
Therefore, while suggestive, the evidence appears strikingly consistent in its avour, even though it
originates from different sources and regulatory areas: large investors believe that there is a clear link
between abuse and high-speed trading techniques. It can almost be read as a call for action on the
part of regulators: for example from the survey SRI commissioned by the Project “Traditional investors
agree that creating a level playing eld and increasing efforts to prevent market manipulation should
be prioritised”
16
.
10 These investors, for example pension and investment funds, are collectively known as the ‘buy side’, following US terminology.
This reects the fact that they ‘buy’ nancial assets with the savings that households entrust to them.
11 The surveys this chapter draws from include: Kay (2012); Finansinspektionen (Swedish Government) (2012); SR1 (Annex D refers).
12 Kay (2012), p. 29.
13 See SR1, p. 6 (Annex D refers).
14 Finansinspektionen (2012), p. 3.
15 IOSCO (2011).
16 SR1, p. 86 (Annex D refers).
92
Market abuse and computer-based trading
More recently still, exchanges and high frequency trading rms have started to acknowledge the issue
publicly. The Chief Executive Ofcer of NYSE Euronext stated that:
The public has never been more disconnected, (…) the citizenry has lost trust and
condence in the underlying mechanism. (…) What used to be an investors’ market
is now thought of as a traders’ market (…) We convinced ourselves (…) that speed is
synonymous with quality and in some cases it might be. In other cases, it clearly isn’t
17
.
It should be noted, however, that more specic interpretation of such qualitative evidence is difcult,
as there seems to be no clear agreement on the specic forms of abuse that might occur. The abusive
strategies evoked include ‘quote stufng’, ‘order book layering’, ‘momentum ignition’, ‘spoong’, and
‘predatory trading’ (sometimes referred to as ‘liquidity detection’ or ‘order anticipation’)
18
. SR1 observes:
Respondents largely agree that it is difcult to nd evidence for market abuse or quantify
its impact on their trades
19
.
There is also a lack of agreement on what regulators should actually do. From the Kay report again:
While the tone of submissions was hostile to HFT, there were few suggestions as to
how the volume of such trading might be reduced or the activities of high frequency
traders restricted
20
.
And again from SR1:
Views diverged considerably across the investor community as to which policy options
are appropriate
21
.
End investors are aware that there is a trade-off between curbing HFT abusive practices and
discouraging legitimate computerised liquidity supply. Therefore the evidence on the perception of
abuse seems compelling in its consistency across sources, perhaps a sign that something potentially
damaging could be at work. But it is also less than helpful to policy makers because agreement is much
weaker when it comes to dening precisely what is wrong or what measures regulators ought to put
in place.
5.3.2 Interpretation: three scenarios
The possibility for elevated abuse arising from HFT activities is consistent with three scenarios,
where the reality or the perception of abuse can be affected. Signicantly, these scenarios are not
mutually exclusive.
In Scenario I, high frequency traders take advantage of slower agents through a speed advantage.
This may involve several of the strategies mentioned above. This corresponds to the claims consistently
emanating from the buy side. It is worth noting that:
• most of these strategies are not new and do not require low latency as such (the obvious exception
being quote stufng, whereby a rm oods electronic systems with order messages to swamp other
participants’ systems with noise), but they may be more protable using speed;
• for the most part, these strategies would be unambiguously classied as manipulative whether
under the EU’s Market Abuse Directive or its successors the Directive on Criminal Sanctions and
the Market Abuse Regulation, since they quite clearly create ‘false markets’, ‘false impressions of
the demand for and supply of securities’, and so on. Legislators, least in current draftings of the new
European directives, have indeed taken the view that there is no obvious need to adapt existing
denitions of abuse to an HFT context.
17 Sloan (2012).
18 Explanations of market abuse strategies can be found in Annex C.
19 SR1, p. 71 (Annex D refers).
20 Kay report (2012), p. 29 op. cit.
21 SR1, p. 25 (Annex D refers).
93
The Future of Computer Trading in Financial Markets
Under Scenario II, the growth of HFT has changed the trading environment, in particular the nature
of the order ow, in ways that facilitate market abuse or increase the perceived extent of abuse. This
covers situations which are different to the confrontation of ‘fast versus slow agent’ described above,
where, for example, high frequency traders prey on each other and slower traders may also nd it
easier to conduct abuse.
The widespread claims of ‘predatory trading’ heard from institutional investors could be a prime
example of this scenario
22
. Computer-based order ow may generate shallower and more predictable
order ow that may be easier for a ‘predator’ (slow or fast) to manipulate because price pressure
effects are easier to exert. This behaviour, especially if conducted through liquidity supply by
temporarily modifying quotes following detection of sustained attempts to buy or sell the asset, could
also be considered manipulative, as it attempts to prot from articially pushing prices away from the
levels warranted by their ‘true’ demand and supply. This statement illustrates:
You can press the button to buy Vodafone, say, and have it executed in a second but in
that period 75% of the liquidity has disappeared and the price has moved
23
.
Scenario II may comprise other situations that can be generated by HFT, which have implications
for abuse but do not in themselves constitute abuse. Rather, they are effects brought about by high
frequency traders by dint of sheer speed. Among those is the HFT-related data ‘overload’ that almost
certainly makes market surveillance vastly more difcult. Other examples are given in DR20.
Finally, Scenario III is one in which other market developments that have accompanied the growth in
HFT (but only partly or not at all brought about by it), may also have contributed to an increase in
the perceived or actual prevalence of abuse. These confounding factors are part of the new trading
environment and may be hard to disentangle from HFT empirically, even though they are conceptually
very distinct.
A prime example of this may be the fragmentation of order ow. Other examples include reduced
pre-trade transparency tied to an increased use of ‘iceberg orders’ and dark venues, and the
exceptionally uncertain economic environment since late 2007 which has led to high volatility and
low liquidity. From SR1:
The macroeconomic environment is seen as the key instigator in the decline in liquidity
24
.
These developments contribute to a sense of unfairness and alienation from the trading process
and, in some cases, may also facilitate abuse. Other illustrations are given in DR20.
The rst Scenario is the one that generates the most speculation but Scenarios II and III may be
equally important.
Can HFT make abuse less likely? This view rarely seems to be entertained, but high frequency
arbitrageurs have powerful means and incentives to keep prices in equilibrium and to inict losses
on manipulators, as they focus on the relative pricing of economically similar securities.
5.4 Future: how might the relationship between computer-based trading and
market abuse evolve in the next ten years?
Concerns have been consistently expressed about the extent of abuse perceived to be conducted
through computer-based techniques, and these perceptions may have damaging consequences for
market functioning. Because we do not know the true extent of abuse, they should not be dismissed
as divorced from reality. What are the available courses of action?
94
22 Similar or identical strategies are called “liquidity detection” and sometimes “front-running” although the latter expression is
inadequate in this context.
23 Paul Squires, head of trading for AXA Investment Managers, (The Economist) 25 February 2012.
24 SR1, p. 51 (Annex D refers).
Market abuse and computer-based trading
5.4.1 Leave it to market mechanisms
It is the case that ‘defence mechanisms’ are being developed by the private sector to address
perceptions of market abuse by high frequency traders. In the last quarter only, ingenious new trading
systems have been launched specically for institutional investors worried about predation, gaming and
other strategies. These systems may feature one form or another of reduced pre-trade transparency,
or they may reintroduce non-anonymity in trading, thus allowing a large investor to select the set of
counterparties they wish to interact with, at the exclusion of, for example, proprietary traders. Other
‘solutions’ include software that assesses the ‘toxicity’ of order ow on each venue by examining
patterns in recent activity and giving a ‘green light’ when trading is considered safe
25
. It has also been
argued that dealing with HFT-driven abuse could be left to trading venue operators themselves, as
they should have strong commercial incentives to provide order books that are seen as ‘clean’.
However, such a purely market-driven approach may not adequately address perceptions of abuse for
at least two reasons. First, many large investors doubt that trading venues have adequate incentives to
be tough with high frequency traders, as HFT rms are one of their most important customer groups.
Leaving it to trading venue operators to police abuse thus runs the risk of harming price formation
by pushing investors to trade ‘over-the-counter’ (OTC) or in dark venues. This in turn increases the
fragmentation of order ow across venues and aggravates order ow imbalances and volatility. Second,
abuse is now possibly easier to conduct across venues and borders than in the past. Thus, market
surveillance that puts too much reliance on individual trading venues is very unlikely to be effective.
As a trader notes:
we monitor the use of our infrastructure to prevent market abuse. But a high-frequency,
multi-broker, multi-exchange strategy could trade buy orders with us, for example, and
the sell orders elsewhere. If we don’t see the sell order, it is very difcult for us to identify
that instance of market abuse
26
.
5.4.2 Address perceptions of the incidence of abuse
This strategy may cover two courses of action that are under active consideration in other mature
markets and ought not, in principle, to be controversial. The details of implementation and the
investment required are not trivial, however.
The rst course of action is to convince market participants that regulators have the ability to deter
abusers. This has two aspects: regulators must be able to levy penalties that are nancially signicant
and they must also convince participants that the probability of abuse being detected and nes being
incurred is not negligible
27
. Regarding the rst aspect, the imposition of large nes in a few high-
prole cases does not seem to have convinced investors that abusers have been deterred. Regulators
seem determined to move on to the next level: making abuse a crime. This is most obviously the
case in Europe under the proposed Directive on Criminal Sanctions for Insider Dealing and Market
Manipulation. As for the second aspect, the market must also be convinced of the regulators’ ability
to detect abuse. The qualitative evidence available suggests that there is considerable room for
improvement, as large investors express a lack of trust in the ability of regulators to tackle abuse.
From SR1:
About 90% of respondents do not believe that regulators have sufcient data, technology
or expertise to effectively detect market abuse
28
.
25 Order ow on a venue may be described as ‘toxic’ when, if a market participant executes on that venue, the probability
of incurring a large trading loss is high. The participant might incur such losses when trading counterparties have superior
information or are able to anticipate the individual’s trading activity.
26 Holley (2012).
27 For a summary of the economics of corporate crime, including market manipulation, see The Economist (2012).
28 SR1, p. 24 (Annex D refers).
95
The Future of Computer Trading in Financial Markets
This in turn may comprise at least two areas for improvement:
• The generation and storage of data that would permit both real-time surveillance and ex-post
investigations, and allow identication of rms and, in some cases, clients. This is being discussed at
European level in continuing debates on transaction reporting, and in the USA, under the perhaps
more ambitious header of ‘Consolidated Audit Trail’ (see EIA17 for more details)
29
. Requiring HFT
rms to keep their trading code for some period, including a description of the rationale behind it,
could also enable better enforcement.
• Once the data are available, there is a need to increase the regulators’ ability to process and
interpret them. In the USA, a report by the Boston Consulting Group recently warned that
the Security and Exchange Commission (SEC) had to substantially improve its sophistication in
information technology to stand any chance of detecting market abuse
30
. The report notes that:
The agency does not have sufcient in-house expertise to thoroughly investigate the inner
workings of such [trading] algorithms (…) [and recommends that] The SEC should have
staff who understand how to (…) perform the analytics required to support investigations
and assessments related to high-frequency trading. (…) the SEC requires people who
know how high-frequency traders use technology
31
.
These issues affect EU regulators to the same extent. Individual brokers can contribute to market
surveillance (by, for example, lling out Suspicious Transaction Reports (STR) introduced in the EU
by the Market Abuse Directive of 2003). One level up, so can exchanges and trading venues, which
operate a mix of real and delayed time abuse/manipulation surveillance systems. As discussed above,
exchange-level surveillance will be particularly useful with respect to strategies that are focused on a
single order book, for example quote stufng or order book layering
32
. But it seems generally accepted
that the hardest part of the task falls to regulators, who must implement monitoring and surveillance
across classes of assets, across exchanges and across borders. This approach to curbing abusive
practices entails both international coordination and substantial investment but these efforts should
both reassure large investors and deter potential abusers.
The second course of action to affect perceptions of abuse is to correct or conrm them through the
production of statistically signicant empirical evidence. HFT activity leaves an electronic trail so, given
the appropriate data, researchers could generate at least preliminary evidence on the prevalence of
simple patterns consistent with abuse. Such evidence could also inform policy making, as the three
scenarios discussed earlier may have very different regulatory implications. At one end of the spectrum,
if abuse is chiey conducted by high frequency traders taking advantage of slower agents, then ways of
slowing down markets may be considered. At the other end, if abuse is largely a matter of perceptions
of large investors who feel alienated from market processes as a result of a new trading environment
made up of speed, fragmentation and/or reduced transparency, then perceptions should be changed.
More large-scale statistical evidence will both help with perceptions and guide regulatory action.
As SR1 concluded:
any regulation of AT or HFT should not be undertaken without substantially greater
analysis and a better understanding of the issues by policy makers and regulators. Some
believe that extensive research is required to determine whether there actually is a
problem in the market stemming from AT or HFT
33
.
29 Very difcult technical issues are involved. An example is to synchronise clocks to the microsecond across the systems of
different electronic venues to ensure that orders and trades can be reliably sequenced, a pre-requisite for any investigation
of trading patterns. See EIA17 (Annex D refers).
30 Boston Consulting Group report (2011).
31 Ibid. p. 53; p. 260.
32 The London Stock Exchange disciplined one of its members for order book layering in 2009. See LSE Stock Exchange
Notice N33/09.
33 SR1, p. 27 (Annex D refers).
96
Market abuse and computer-based trading
5.5 Conclusions
Economic research thus far provides no direct evidence that high frequency computer-based trading
(CBT) has increased market abuse, although this research is at an early stage and incomplete: its main
focus is not on the measurement of market abuse during the continuous phase of trading. However,
claims of market manipulation using HFT techniques are consistently reported by institutional investors
(pension funds or mutual funds) internationally. While these claims are not substantiated by evidence,
plausible scenarios can be constructed which show how such abuse could potentially occur.
Policy makers should take such perceptions seriously, whether or not they are true: the actual
extent of abuse can never be known and it is perceptions that determine trading behaviour and
investment decisions.
Currently, surveys show that institutional investors have little or no condence in the ability of
regulators to curb the behaviour that those investors describe as abusive. While market mechanisms
will provide some help, it is likely that these concerns will need to be addressed more directly. This
chapter has argued that the need for direct intervention in market operation is not demonstrated
at this stage, but that increasing the ability of regulators to detect abuse and produce statistically
signicant empirical evidence on its extent would help either to conrm concerns or lay them to rest,
and thus restore market condence.
The Report has considered a number of aspects related to CBT and outlined some of the challenges
that CBT presents. In the next chapter, the report considers policy measures that aim to address those
challenges and what economic impact those measures might have.
97
6 Economic impact assessments
on policy measures
Key findings
Computer trading has changed markets in fundamental ways, not the least of which is the speed at
which trading now occurs. There are a variety of policies proposed to address this new world of
trading with the goals of improving market performance and reducing the risks of market failure.
These policies include notication of algorithms, circuit breakers, minimum tick size requirements,
market maker obligations, minimum resting times, minimum order-to-execution ratios, periodic
auctions, maker-taker pricing, order priority rules and internalisation. The Foresight Project has
commissioned a variety of studies to evaluate these policies, with a particular focus on their economic
costs and benets, implementation issues and empirical evidence on effectiveness. This chapter
summarises those ndings.
The key ndings relating to the different policies are as follows, starting with those which were most
strongly supported by the evidence:
• Overall, there is general support from the evidence for the use of circuit breakers, particularly for
those designed to limit periodic illiquidity induced by temporary imbalances in limit order books.
Different markets may nd different circuit breaker policies optimal, but in times of overall market
stress there is a need for coordination of circuit breakers across markets.
• There is also support for a coherent tick size policy across similar markets. Given the diversity of
trading markets in Europe, a uniform policy is unlikely to be optimal, but a coordinated policy across
competing venues may limit excessive competition and incentivise limit order provision.
• The evidence offers less support for policies imposing market maker obligations. For less actively
traded stocks, designated market makers have proven benecial, albeit often expensive. For other
securities, however, market maker obligations run into complications arising from the nature of
high frequency market making across markets, which differs from traditional market making within
markets. Many high frequency strategies post bids and offers across correlated contracts.
A requirement to post a continuous bid-offer spread is not consistent with this strategy and, if
binding, could force high frequency traders out of the business of liquidity provision. Voluntary
programmes whereby liquidity supply is incentivised by the exchanges and/or the issuers can
improve market quality.
• Similarly, minimum resting times, while conceptually attractive, can impinge upon hedging strategies
which operate by placing orders across markets and expose liquidity providers to increased ‘pick-off
risk’ if they are unable to cancel stale orders.
• The effectiveness of the proposed measure to require notication of algorithms is also not
supported by the evidence. The proposed notication policy is too vague, and its implementation,
even if feasible, would require excessive costs for both rms and regulators. It is also doubtful that it
would substantially reduce the risk of market instability due to errant algorithmic behaviour, although
it may help regulators understand the way the trading strategy should work.
• An order-to-execution ratio is a blunt policy instrument to reduce excessive message trafc and
cancellation rates. While it could potentially reduce undesirable manipulative trading strategies,
benecial strategies may also be curtailed. There is insufcient evidence to ascertain these effects,
and so caution is warranted. Explicit fees charged by exchanges on excessive messaging and greater
regulatory surveillance geared to detect manipulative trading practices may be more effective
approaches to deal with these problems.
99
The Future of Computer Trading in Financial Markets
Key findings (cont.)
• The issue of maker-taker pricing is complex and is related to other issues like order routing, priority
rules and best execution. Regulatory focus on these related areas seems a more promising way of
constraining any negative effects of maker-taking pricing than direct involvement in what is generally
viewed as an exchange’s business decision.
• The central limit order book is of fundamental importance and everyone involved in equity trading
has an interest in changes that improve the performance of the virtual central limit order book
currently operating in Europe. The debate on how best to do so continues, in part because the
opposing trends towards consolidation of trading venues and fragmentation is likely to remain
unresolved in the near future.
• Internalisation of agency order ow, in principle, benets all parties involved, especially where large
orders are involved. However, the trend away from pre-trade transparency cannot be continued
indenitely without detrimental effects on the public limit order book and price discovery.
• Call auctions are an alternative trading mechanism that would eliminate most of the advantage
for speed currently present in electronic markets. They are widely used already in equity markets
at open and close and following a trading halt. But no major market uses call auctions exclusively
to trade securities. To impose them as the only trading mechanism seems unrealistic as there are
serious coordination issues related to hedging strategies that make this undesirable.
100
Economic impact assessments on policy measures
6 Economic impact assessments
on policy measures
6.1 Notification of algorithms
6.1.1 The measure and its purpose
Algorithmic trading (AT) involves the use of computer programs to send orders to trading venues.
Such algorithms now have widespread use among all classes of investors
1
, and AT comprises the bulk
of trading in equity, futures and options markets. AT is also fundamental to high frequency trading
(HFT) strategies. A concern with AT is that an errant algorithm could send thousands of orders
in milliseconds to a market (or markets), resulting in major market upheaval. Markets in Financial
Instruments Directive (MiFID) II Article 17(2) proposes that investment rms engaging in AT must
provide annually to the regulator a description of their AT strategies, details of the trading parameters
and limits, the key compliance and risk controls which are in place, and details of how its systems are
tested. The purpose of this measure is to ensure that algorithms used in trading are subject to proper
risk controls and oversight.
6.1.2 Benefits
If descriptions were able to prevent unsound algorithms from operating in live markets, then this
would be a measure contributing to the maintenance of orderly markets. Requiring rms to have
demonstrated risk controls in place might make aberrant algorithms less likely to occur. If regulators
require increased testing at the rm level for algorithms it suspects are awed, fewer events affecting
market liquidity due to malfunctioning algorithms might occur.
One additional benet is that regulators will have to acquire greater technical sophistication to
understand and evaluate the algorithms being used in trading, which would improve their ability to
investigate abusive practices. However, this would require substantial increases in personnel and
greater investments in technology.
6.1.3 Costs and risks
There are substantial costs connected with meeting notication requirements in general, and
particularly as currently stated in Article 17(2). Cliff (EIA16)
2, 3
argues that just providing a full
description of an AT strategy requires not only all the programs that have been written to implement
it, but also the full details of the code libraries used, as well as the software tools involved. Moreover,
these descriptions must include the actual computations required, the algorithms that affect the
computations, and full details of how the algorithms are implemented. Providing information on the
other aspects of the proposed regulation would be similarly complex. Regulators, in turn, would then
have to analyse this material, and determine what risk, if any, this algorithm poses to the market. This
would require substantial expertise at the regulator level with complex computing systems and analysis.
An additional risk to consider is that algorithms are updated frequently, meaning that annual reviews
will be ineffective in actually capturing the risk facing the markets.
1 The survey SR1 (Annex D refers) commissioned by the Foresight Project found that algorithmic trading is used by 95% of asset
managers, 100% of insurers and 50% of pension funds surveyed.
2 Throughout this document EIA refers to economic impact assessment studies commissioned by the Project. These can
be found in the Project’s webpage: http://www.bis.gov.uk/foresight/our-work/projects/current-projects/computer-trading/
working-paper Accessed: 17 September 2012.
3 EIA16 (Annex D refers).
101
The Future of Computer Trading in Financial Markets
Cliff (EIA16) estimates that it could run to approximately one billion Euros a year if the descriptions
were in fact carefully read. Alternatively, the cost of Article 17(2) could be dramatically lowered by
simply having rms provide documents to the regulator that are led but not really analysed. In this
case, however, it is hard to see how this activity could actually address the potential risk of algorithmic
disruptions to the market.
Care must be taken not to infer that this measure would dramatically reduce systemic risks, and that
agents would consequently take larger risks than they otherwise would have. The reason systemic risk
may not be reduced signicantly even if algorithms were carefully analysed by regulators is that much of
the risk arises from the nonlinear interactions of many algorithms. Different algorithms may be present
in the markets at different times, setting up what could be innite combinations of algorithms to
consider for regulatory review. Furthermore, even if a ‘wind tunnel’
4
for testing algorithmic interaction
were constructed, it would not capture all of the systemic risks if algorithms learn and rewrite
themselves dynamically over time.
6.1.4 Evidence
There is very little empirical evidence on the cost or benets of algorithm notication. Cliff (EIA16)
provides some estimates of cost, but these depend greatly on how the notication requirements are
implemented and on how much the required analysis is split between rms and the regulator. There
is to our knowledge no cost estimate of risk controls for algorithms at the rm level.
6.1.5 Conclusions
The desirability of understanding AT strategies and their impact on the market is laudable but achieving
this through notication requirements of the type currently envisioned in MiFID II may not be feasible
given the complexity of algorithms and their interactions in the market.
6.2 Circuit breakers
6.2.1 The measure and its purpose
Markets have always been subject to episodic price instability, but computerised trading combined with
ultra low latency creates increased potential for such instability to occur. This, in turn, has increased
interest in the role and usage of circuit breakers. Circuit breakers are mechanisms for limiting or halting
trading on exchanges. Their purpose is to reduce the risk of a market collapse induced by a sequence of
cascading trades. Such trading could arise from pre-specied, price-linked orders (such as programme
trades or trade algorithms that sell more when prices fall), or from self-fullling expectations of falling
prices inducing further selling. Traditionally, circuit breakers were triggered by large price movements,
and hence represented ex-post reactions to excessive price volatility in the market. More recently,
the advent of HFT taking place at millisecond speeds has resulted in a new generation of circuit
breakers which work on an ex-ante basis (i.e. halting trading before accumulated orders are executed).
Circuit breakers can take many forms: halting trading in single stocks or entire markets; setting limits on
the maximum rises or falls of prices in a trading period (limit up and limit down rules); restrictions on
one trading venue or across multiple venues. The London Stock Exchange (LSE), for example, operates
a stock-by-stock circuit breaker that, when triggered, switches trading to an auction mode in which
an indicative price is continuously posted while orders are accumulated on either side. After some
time, the auction is uncrossed and continuous trading resumes. The trigger points are in several bands
depending on the capitalisation and price level of the stock, the recent transaction history and the most
recent opening prices.
102
4 What is meant here is a simulated market that can in principle test algorithmic interaction much as a wind tunnel is used for
testing aircraft designs.
Economic impact assessments on policy measures
However, circuit breakers are no panacea. Price discovery is a natural feature of markets, and bad news
can induce (sometimes large) price drops to new efcient values. Halting markets can interfere with
this natural process, and may simply postpone the inevitable. For example, the October 1987 crash
in the USA was subsequently followed around the world, with big price drops in the UK and other
markets. The Hong Kong stock market was closed on the Monday after the US markets had started to
fall and stayed closed for a week. When it did open, it suffered an instant decline of 30%. On the other
hand, the May 2010 Flash Crash in New York was effectively ended by a circuit breaker which allowed
liquidity to re-accumulate as buyers returned to the market and the newly balanced market to resume.
Thus, circuit breakers, while well suited to dealing with instability caused by temporary shortages of
buyers or sellers, are not appropriate for all causes of market volatility and cannot forestall revaluations
that are unavoidable.
The policy issue is whether the existing self-regulatory, uncoordinated approach (in Europe) to price
instability can be improved.
6.2.2 Benefits
If the price processes are driven purely by rational valuations of fundamentals, then a trading halt
impairs the process of valuation and prevents the public from receiving accurate and up to date
information. But in today’s high frequency electronic markets, liquidity is uncertain, and prices can be
affected (at least temporarily) by factors such as imbalances in the book of orders, fat nger trading
errors, and errant algorithms engaged in a mechanical feedback loop. Circuit breakers and trading halts
may be benecial for dealing with such imperfections in trading processes. We describe the benets of
circuit breakers in the following three subsections.
Cooling-off period
Many modern markets function as computerised limit order books with continuous trading and
replenishment of orders. Even if the daily trading volume is large, the displayed depth of the market at
any moment may be relatively small. A large purchase order arriving unexpectedly, for example, can
cause a temporary imbalance until more sellers come forward. Circuit breakers provide a respite that
prevents mechanical selling at any price, allows the market to understand what is happening and gives
counterparties time to enter, thereby reducing the order imbalance.
In a fast-moving market, losses to positions bought using credit can build up quickly, leading the brokers
who have provided credit to ask for additional collateral. With very fast-moving markets, these margin
calls cannot be satised quickly enough and broker condence may suffer. A cooling-off period allows
the traders to raise the collateral, reducing the risk that they fail. It also reduces market risk because
brokers will not be forced to liquidate clients’ positions which would then put additional selling pressure
on the market and create a vicious feedback loop.
Circuit breakers can also be invoked for purely technical reasons to prevent peak overload bottlenecks
at the exchanges’ servers which could lead to erroneous pricing and execution.
Uncertainty resolution
Volatility is a natural part of markets, but unexplained volatility can cause traders to fear the worst
and lead to massive selling. Because high frequency markets move so fast, it may be impossible for
traders to evaluate what is causing a large price movement. There is now a large and growing literature
showing that uncertainty reduces participation in markets, which can manifest in massive selling for
those in the market and a reluctance to participate by those not already there. Either outcome is
undesirable, so mechanisms such as circuit breakers that can allow time for uncertainty resolution can
be benecial.
103
The Future of Computer Trading in Financial Markets
Investor protection
A market that is fast-moving for reasons other than the inow of fundamental news can not only
create systemic risk but can also penalise traditional investors who do not have the resources to
monitor markets continuously. Trading halts in response to non-fundamental swings may offer a means
of preventing uninformed retail investors losing out to traders who continuously monitor markets.
This could bolster the condence of investors in the integrity of markets, and remove or ameliorate
concerns that small investors can be taken advantage of by manipulative trend-generating strategies
5
.
A nal point is that circuit breakers enjoy widespread support from industry as shown in the survey
SR1. They are seen as a prudent mechanism to enforce orderly, fair and efcient markets.
6.2.3 Costs and risks
The obvious cost is that during a market halt traders are prevented from completing mutually benecial
trades. An empirically documented effect of circuit breakers is the so-called ‘magnet effect’ whereby
traders rush to carry out trades when a halt becomes imminent, accelerating the price change process
and forcing trading to be halted sooner or moving a price more than it otherwise would have moved.
Subrahmanyam (EIA4)
6
recommends that there should be some unpredictability about when circuit
breakers are triggered, a logic that may explain Deutsche Börse’s decision not to publish the trigger
points for its circuit breakers.
Similarly, a trading halt that slows down the fundamental price discovery process may create additional
uncertainty. This can undermine condence and increase bid-ask spreads when markets reopen if the
reason for the halt is not credible. More generally, if circuit breakers are implemented across trading
venues and coordinated so that a trading halt for a stock on one venue triggers a trading halt for
that stock on all other venues, then a halt on a minor trading facility can trigger an unexplained halt
on the main venue. In fact, experimental evidence suggests that completely random halts can create
uncertainty because traders have the time to make up rumours or to exaggerate existing ones
7
.
An additional detrimental element associated with circuit breakers is the inability for market makers
to ofoad large positions quickly when trading is halted. This would have to be factored into their risk
management system and might make them less willing to buy, reducing liquidity.
Similarly, there is an issue with respect to cross-asset, cross-venue trade. Suppose a trader is hedging
a derivative or is engaged in arbitrage across securities. If trading in the stock is halted, the trader
is suddenly no longer hedged and may suffer losses as a result. If this happens frequently, different
markets may become less integrated, market efciency suffers, and the possible loss of condence
in pricing accuracy can lead to feedback loops. If investors suddenly lose condence in the price of a
security (because trading has stopped) and they are invested in other assets in, perhaps, a carefully
designed portfolio, they may decide to sell many other assets because their ability to control the total
risk they face is compromised. This may lead to a chain of negative events across many securities.
A mandatory, market-wide circuit breaker has not existed in the UK, although in extreme circumstances
the LSE declares a ‘fast market’ and changes certain trading rules (for example, widening individual
stock circuit breakers and relaxing market maker obligations). Due to the nature of HFT, much more
trading now involves the futures market, exchange traded funds (ETFs), contracts for difference (CFD)
and spread betting
8
. The US Flash Crash was triggered in the S&P500 E-Mini futures market, then went
to the ETFs on the index, and nally affected the equity market itself.
104
5 Kim & Park (2010).
6 EIA4 (Annex D refers).
7 EIA9 (Annex D refers).
8 In the UK, CFDs are often used as an indirect way of trading equities but avoiding the stamp duty on equity trading, since
CFDs are not covered by this tax.
Economic impact assessments on policy measures
This raises the problem of how to implement circuit breakers across market venues. In the US Flash
Crash, attempts by the NYSE to slow trading were completely ineffective due to the ability of traders
to use other venues that were not affected. For a uniform circuit breaker to be effective, it would have
to close all markets for a single stock or series of stocks. This would require coordination between and
across exchanges, but different exchanges have different trading rules and different trading practices.
Moreover, the ability of an exchange to determine its own rules for handling volatility can be viewed as
an important dimension of its risk management practices. Regulators would have to consider whether
it is desirable for an erroneous trade on, for example, Chi-X to shut down the LSE, or whether the
current system whereby only the primary market determines shutdowns is preferable
9
.
A hybrid system allowing regulatory override of individual exchange trading halt decisions might
provide a mechanism to deal with market-wide disturbances.
6.2.4 Evidence
There is a sizeable literature on the impact of circuit breakers on markets, but overall the empirical
results are mixed. Many authors nd negative effects of circuit breakers, while others nd no effects or
small positive effects. However, it is difcult to analyse what would have happened had the market not
been halted, and so with a few exceptions these ndings are not statistically robust. A more important
problem is that high frequency markets are very different from the markets analysed in previous
research. There is, as yet, little academic research on the role of circuit breakers in high frequency
market settings. In particular, many academic studies are on smaller international markets and are
largely concerned with a simple type of breaker rather than the more complex models employed
for instance by the LSE
10
. Furthermore, most analyses have almost exclusively focused on transaction
prices in the immediately affected market rather than the order book or the spill-over effects on other
securities and trading venues.
There is some evidence from the US Flash Crash on the effectiveness of modern circuit breakers.
The end of the crash is generally attributed to the imposition of the Chicago Mercantile Exchange’s
‘Smart Logic’, a circuit breaker that halts trading when the accumulated imbalance of pending orders,
if executed, would result in the price falling beyond a pre-specied limit. This forward-looking circuit
breaker differs from the variety generally employed in most markets which deal with issues once they
arise, and may provide a template for the design of market-wide circuit breakers.
There is also some evidence on the use of single stock circuit breakers. The LSE describes how on
an average day there are 30–40 trading suspensions, whereas in the rst two weeks of August 2011
(when there was a great deal of volatility), this shot up to about 170 suspensions per day. Despite the
high number of suspensions, large volume and wide market swings, trading was generally ‘orderly’ in
their view
11
. Of course, there are other market-wide costs associated with each stoppage and it is not
clear whether these costs are fully taken into account and appropriately balanced against the benets.
6.2.5 Conclusions
Circuit breakers have a role to play in high frequency markets, and they are found in virtually all
major exchanges. Because of the inter-connected nature of markets, however, there may be need
for coordination across exchanges, and this provides a mandate for regulatory involvement. New
types of circuit breakers which are triggered before problems emerge rather than afterwards may
be particularly effective in dealing with periodic illiquidity.
9 The costs and benets of coordinated trading halts are further analysed in EIA20 (Annex D refers).
10 A notable exception is a study by Abad & Pascual (2007) who investigate the type of circuit breaker used on the London
Stock Exchange and the Spanish Stock Exchange.
11 The London Stock Exchange Group (2011).
105
The Future of Computer Trading in Financial Markets
6.3 Minimum tick sizes
6.3.1 The measure and its purpose
The minimum tick size is the smallest allowable increment between quoted prices in a market.
Tick sizes have important implications for both transaction costs and liquidity provision. The transaction
cost effect is straightforward: a larger tick size, if binding, increases trading costs by widening the spread
between bid and offer prices. The liquidity effect arises because the tick size determines how easy it is
for another trader to ‘step ahead’ of an existing limit order. In markets with a standard price/time
priority (PTP) rule, an order placed rst executes ahead of one placed later unless the later order
is posted at a ‘better’ price. Small tick sizes make it easier for traders to post that better price, so
smaller tick sizes push up the cost for traders who put trades on the market and provide liquidity.
The challenge for markets and regulators is to choose the optimal tick size to balance these liquidity
and transaction cost effects. An additional challenge is to decide whether or not minimum tick sizes
need to be harmonised across linked venues.
There are important differences in minimum tick size policy between the USA and Europe. In the
USA, Regulation National Market System (NMS) requires that in all ‘lit’ venues (exchanges and large
Alternative Trading Systems (ATS)) stocks over $1 are quoted with a minimum tick size of one cent,
and sub-penny pricing is prohibited. In Europe, there is no mandated tick size and local exchanges
are free to set their own tick policy. As a result, there is generally a range of tick sizes depending on
the price level and, in the case of LSE, on the market capitalisation. A European stock may trade on
different public venues under different tick size regimes, whereas in the USA, such differences can
only currently happen in dark venues. Historically, the trend has been towards smaller tick sizes since
US trading in ‘eighths’ (12.5 cents) yielded to decimalisation in 2000. Now, active stocks in the USA
typically trade at one cent spreads, leading to concerns that a one cent minimum tick may be too large
thus illustrating the earlier mentioned trade-off between transaction costs and liquidity provision.
In Europe, spreads at minimum levels are not as common, suggesting that the tick rules are not as
binding on market behaviour. The policy issue is whether it would benet market quality to mandate a
minimum. A related policy issue is whether minimum tick sizes need to be harmonised across venues.
6.3.2 Benefits
Choosing an optimal minimum tick size for a given stock and a given market environment would have
a number of benets
12
. Originally, a uniform tick size rule was chosen to minimise transaction costs for
rms and traders. With a limited number of possible prices, the technological complexity of trading
was reduced and also the cost. While this is no longer needed given the state of trading technology, it
is still the case that a well-chosen minimum tick size framework can reduce the need for rms to split
or reverse split their stock in order to inuence the relative tick size. This latter reason speaks to the
benets of a non-uniform tick policy as found in Europe.
A well-chosen minimum tick size can prevent market operators such as exchanges, market makers
or high frequency traders making excessive prots at the expense of the nal users. Chordia (EIA6)
argues that a too high minimum tick in the USA led to payment for order ow and internalisation as
competitive responses to the excessive prots accruing to liquidity providers in large stocks. To the
extent that trading venues compete with each other in a world with HFT and maker-taker pricing
13
,
exchanges may have an incentive to try to undercut each other’s tick sizes to attract volume and hence
fees. A coherent overall minimum tick size policy (such as the one agreed to by Federation of European
Securities Exchanges (FESE) members, discussed below) that applies to all trading venues could prevent
a welfare-destroying race to the bottom where competitive pressures would otherwise lead to a tick
size below the optimal range.
106
12 For further discussion on the optimal tick size and its modelling see EIA7 (Annex D refers).
13 Maker-taker pricing refers to the practice in many exchanges and trading platforms of paying a small rebate to executed orders
that were placed as passive limit orders (the liquidity makers) and charging a fee to active orders that hit existing limit orders
(the liquidity takers).
Economic impact assessments on policy measures
A higher minimum limit for tick sizes can also provide greater incentives to post limit orders and
thereby create a deeper, more liquid market. Because the costs of jumping the queue are higher with
larger minimum spreads, there is less queue-jumping and so a generally more stable book. The reduced
frequency of limit order book updating means less data and lower costs related to data. Whether
such changes would reduce the arms race of trading rms investing in ever faster trading systems is
debatable. A larger tick reduces the incentives to undercut, but it still remains the case that the rst
trader to post gets the order. This would imply that high frequency traders will generally dominate the
order book for active stocks regardless of tick sizes.
Although Europe does not as yet have a mandatory tick size, there have been attempts by the industry
through its organisation FESE to harmonise and simplify the tick size regimes across their members.
There has been some agreement, but there is no binding legal framework to enforce it and it may not
prove sustainable over the long term in the presence of new entrants.
6.3.3 Costs and risks
Setting too large a minimum tick size results in large bid-ask spreads favouring market makers. If prices
must be set ve pence apart, for example, then the bid-ask spread can also be no smaller than ve
pence. Except in special cases, a spread of one tick size cannot be undercut because prices must be
quoted in increments of the tick size. A large tick size also reduces the chance of a price improvement.
A price improvement occurs in any auction when a higher price is offered for the good. If the only bids
accepted are in £1,000 increments it is much harder to improve the price than if you are allowed to
raise your offer by £1.
An excessive tick size on one market may induce activity to migrate towards venues with lower ticks.
In the USA, this is one reason given for the growth of trading in dark venues, and in Europe such tick
size competition might be expected to induce orders to ow towards dark or systematic internaliser
(SI) venues that are not subject to the rules for multilateral trading facilities (MTF). This migration is
most likely to happen with low-priced stocks where the bid-offer spread is large as a proportion of
the stock’s value. Migration can be mitigated through additional measures, such as coordinated rules
across trading venues and SIs, or via mechanisms such as the retail liquidity providers (RLPs) proposed
by the NYSE and recently approved by the US Securities and Exchange Commission (SEC)
14
. These
RLPs would be allowed to quote on sub-penny ticks provided their quotes are hidden and can only be
accessed by retail order ow.
Setting too small a minimum tick size comes with costs as well. In markets with PTP, passive traders
submit limit orders and thereby give away a free option to market participants. These passive traders
expect to cover these losses through the spread. If tick sizes are very small, then new passive traders
can come in and capture the incoming marketable orders by undercutting the current best bid or ask
by a tick. Traders who provide liquidity by putting trades on the market will not be rewarded for having
taken the risk of being picked off by traders with new information or through adverse selection. Too
small a cost for jumping the queue, therefore, makes providing visible liquidity more expensive and
leads to smaller and more ephemeral depth. It may also contribute to more cancellations and eeting
limit orders in the book as traders try to snipe in as late as possible.
14 The US SEC approved the NYSE proposed pilot on July 3, 2012 (2012a).
107
The Future of Computer Trading in Financial Markets
6.3.4 Evidence
There is a large academic literature investigating the inuence of tick sizes on market behaviour
15
.
In general, the results from studies of a wide range of markets nd that a reduction in tick sizes
reduces spreads but also reduces depth. As a result, transaction costs for smaller retail investors tend
to be lower, but the results are ambiguous for institutional investors whose trades are in sizes that
may require greater depth. These empirical ndings are consistent with the recent pilot programme
implemented by FESE and subsequently analysed by Better Alternative Trading System (BATS) (2009)
16
.
Citicorp’s recent reverse stock split underscores the effects that changing the relative tick can have
on trading
17
. After Citicorp substituted one share for ten, trading in its stock shifted from alternative
trading venues to exchanges, HFT activity increased, the trade size rose on alternative venues, volatility
was higher and volume lower, and interestingly, order ow toxicity was lower
18
. What is not yet
established are the effects on traders or on the issuer.
While the literature carefully delineates the impact of tick size changes on traders and markets, there
is no proper analysis on who should make the tick size decision and whether the minimum tick size
should be the same for all rms or all trading venues. The current approach in Europe of allowing
each venue to choose its own minimum tick size has a variety of merits, but it can lead to excessive
competition between venues. The US approach of a uniform minimum tick size removes this problem,
but there are deep concerns that it is leading to insufcient liquidity for less frequently traded stocks
19
.
6.3.5 Conclusions
Tick size policy can have a large inuence on transaction costs, market depth, and the willingness to
provide liquidity. The current approach of allowing each European trading venue to choose its own
minimum tick size has merits, but can result in unhealthy competition between venues and a race to
the bottom. A uniform policy applied across all European trading venues is unlikely to be optimal, but a
coherent overall policy for minimum tick size that applies to subsets of trading venues may be desirable.
This coordinated policy could be industry-based, such the one agreed to by FESE members.
6.4 Obligations for market makers
6.4.1 The measure and its purpose
Obligations for market makers are requirements that a person (or more controversially, a computer
program) acting as a market maker must post prices to buy and sell at competitive levels at all times
the venue is open and regardless of market conditions. This could be applied to traditional (human)
market makers or to algorithmic market makers or both
20
. The purpose of this proposal is to improve
continuous liquidity provision and to ensure that market makers are actively quoting competitive prices
during periods of market stress.
Market makers provide liquidity to traders by being willing to buy when a trader wants to sell and to
sell when a trader wants to buy. For providing this service, the market maker earns the bid-ask spread.
In the case of large, actively traded issues held by both institutions and retail traders, market makers
typically earn sufcient returns to justify the time and capital needed to perform this function. For
less actively traded stocks, this may not be the case, and wide spreads, reduced depth and general
illiquidity may result. In times of market turmoil, market making in any stock is often unprotable and
the withdrawal of market makers can result in market-wide illiquidity. Consequently, market makers
15 See EIA22 for a survey (Annex D refers).
16 See BATS Trading Ltd (2009). London Economics have calculated the economic impact of certain tick size changes through the
effect this would have on bid-ask spreads and hence on the cost of capital. See EIA22 (Annex D refers).
17 ITG ( 2011).
18 Order ow is toxic when it adversely selects market makers, who may be unaware they are providing liquidity at a loss. The
fall in order ow toxicity suggests that adverse selection also fell after the reverse stock split.
19 A further issue, as discussed by Angel (EIA7), is that suboptimal tick sizes may lead to stock splits or reverse stock splits by
companies who have a different view of what the tick size should be in relation to the price. This splitting phenomenon seems
to be more common in the USA, which has a relatively rigid tick size policy.
20 The current MiFID II draft mentions such a proposal in Article 17(3).
108
Economic impact assessments on policy measures
have often received various inducements to supply liquidity such as access to fee rebates, or superior
access to the order book, or exemption from short sale requirements, or even direct payments from
exchanges or issuers. In return for these inducements, exchanges traditionally required market makers
to quote bid and ask prices even in times of stress.
In current markets, much of the market maker function has been taken over by HFT in which a
computer algorithm is programmed to buy and sell across markets. Such computerised trading often
involves the placement of passive orders (i.e. limit orders) and so, like the traditional market maker, the
HFT program is buying from active traders who want to sell and selling to active traders who want
to buy. However, unlike traditional market makers, the HFT program is not committed to a particular
venue, and generally does not have access to superior information, although the computer running the
programme may be co-located in the exchange. What is now being considered is whether such high
frequency market making should also face obligations with respect to provision of liquidity. Among the
various obligations being considered are: maximum spread restrictions; percentage time for quotes to
be at the inside spread; minimum quoted size; and minimum quote time.
The policy issue is whether regulators should require market maker obligations for any or all types of
market makers. Many exchanges already have some obligations for market makers in terms of quoting,
so the question is whether this needs to be mandated across all trading venues or extended more
broadly to the market making function
21
.
6.4.2 Benefits
Market maker obligations can improve market quality and hence raise social welfare. Narrower
spreads will induce both informed and uninformed traders to trade more, which in turn increases
price efciency and quickens price discovery. To the extent that obligations improve the depth of the
market, traders will nd it easier to buy and sell, and transaction costs should be lower. Obligations to
set competitive prices could help reduce volatility, and requirements to stay in the market continuously
could lead to greater liquidity during periods of market stress.
It should be noted that benets from such obligations are not guaranteed because of the high costs
that they may entail. When companies have contracted for market making services (as, for example, on
the Stockholm Stock Exchange), Weaver (EIA8)
22
reports increased market quality in terms of trading
volume and liquidity measures for their shares. However, the high costs of paying for contracting have
deterred many small companies from doing so.
6.4.3 Costs and risks
Market maker obligations would be unnecessary if providing liquidity was a protable and relatively
riskless undertaking. The reality, however, is far different, and market makers face a variety of situations
in which posting quotes exposes them to the risk of large losses. Moreover, even in the best of
circumstances, market making is not cost free, requiring both capital and expensive investments in
technology to support operations. To the extent that market making obligations are imposed without
corresponding compensation, at least some market makers will exit the market, reducing its liquidity.
Imposing these obligations is problematic. Rules requiring market makers to post narrow bid-offer
spreads are often unnecessary for large, active stocks where market making is protable. However,
with less actively traded, small stocks, the order ow is more naturally unbalanced and the market
maker faces substantial risk acting as the intermediary for the bulk of trading. Market maker obligations
on such stocks will impose signicant costs on market makers, and too little market making will
be provided unless they are compensated. Conversely, if market makers are given too much
compensation, trading in small stocks will essentially be subsidised. Such a situation characterised
trading on the NYSE (before the regulatory changes in the late 1990s) whereby small stocks generally
beneted from rules restricting maximum spreads and mandating continuous prices.
21 For example, NASDAQ has recently proposed the designation of market makers who are compensated by issuers
when committing to a minimum liquidity supply. This new programme is under review. US Securities & Exchange
Commission (2012b).
22 EIA8 (Annex D refers).
109
The Future of Computer Trading in Financial Markets
Market making during times of market stress is also an extremely risky proposition as requirements to
buy when prices are crashing may lead to bankruptcy for the market maker. An optimal market maker
obligation should not force a market maker into bankruptcy, and although limits as to what is actually
required are necessary, they are difcult to even dene, let alone enforce. Obligations which are too
stringent will transfer losses from traders to market makers, while requirements which are too lax can
result in greater market instability.
Imposing market maker obligations on algorithmic market making trading strategies raises a variety
of risks. Many high frequency strategies post bids and offers across correlated contracts. Thus, a
high frequency market maker may be buying in one market and selling in another. A requirement to
post a continuous bid-offer spread is not consistent with this strategy and, if binding, could force high
frequency traders out of the business of liquidity provision. With upwards of 50% of liquidity coming
from high frequency traders, this could be disastrous. A more likely outcome, however, is that any
requirement would be evaded by posting one quote at the market and the other off the market or for
small size
23
. Moreover, what constitutes market making in this context is unclear. Many high frequency
strategies are actually a form of statistical arbitrage, buying where prices are low and selling where
prices are high. Using limit orders to implement these strategies is akin to market making, but it also
differs in a variety of ways. Forcing market maker obligations on algorithmic trading could reduce these
statistical arbitrage activities and thereby reduce market efciency. These issues are discussed in more
detail in Cliff (EIA19).
Cliff (EIA19) also discusses the challenges of specifying a requirement that would apply to algorithmic-
based market making strategies in a logical or enforceable way. Changes to the original MiFID II 17(3)
specication (known as the Ferber amendments) mean that market maker obligations would apply
only to HFT systems that operate on maker-taker trading venues and for which more than 50%, or a
majority, of the system orders/trades qualify for maker discount/rebates. While this revised application
may be more feasible to implement, it may also induce trade to move to venues where this regulation
is not binding.
6.4.4 Evidence
The vast majority of empirical studies on voluntary
24
market maker obligations conclude that they
improve market quality. These benets are found in a wide variety of market settings, and across
different classes of securities such as equities and options. The benets are especially felt in illiquid
stocks, where generating critical mass in a market is an issue. The fact that virtually every major
stock exchange has some form of market maker obligation testies to their usefulness in enhancing
market behaviour
25
.
However, empirical research nds that traders under market maker obligations generate these benets
in part because they get compensated. This can be in the form of extra rights such as sole view of the
limit order book, ability to short shares ‘naked’ (without holding the stock) or direct compensation
paid either by the trading venue or by the listing company. Moreover, there is a surprising diversity of
approaches taken towards how these market maker obligations are structured. EIA8 notes that “the
application of minimum obligations for market makers, as well as the mode of compensation, is uneven
across markets on both sides of the Atlantic. Some markets impose minimum obligations on market
makers for all listed stocks.”
6.4.5 Conclusions
The current system of exchanges determining how to structure market maker obligations and pay
for them seems to be working well for most markets. We think there is less support for policies that
impose market maker obligations for a large class of market participants without a thought-through
incentive scheme to support it.
23 See EIA22 for further discussion of monitoring and enforcement issues (Annex D refers).
24 By ‘voluntary’ we mean that these all refer to situations where there is some quid pro quo for the market making obligation in
terms of fee rebates and superior access to the order ow. Compulsory market making obligations (i.e. without compensation)
have not been studied to our knowledge.
25 Of the major markets, only the Tokyo Stock Exchange does not have some form of specied market making. See Weaver
(EIA8) for a detailed discussion of these issues.
110
Economic impact assessments on policy measures
6.5 Minimum resting times
6.5.1 The measure and its purpose
Minimum resting times specify a minimum time that a limit order must remain in force. The impetus for
imposing a minimum is that markets now feature a large number of eeting orders that are cancelled
very soon after submission. This increases the costs of monitoring the market for all participants, and
reduces the predictability of a trade’s execution quality since the quotes displayed may have been
cancelled by the time the market order hits the resting orders. The nature of high frequency trading
across markets, as well as the wide-spread usage of hidden orders on exchanges, are responsible for
some of this eeting order behaviour. However, frequent cancelling of quotes may also result from
abusive strategies including spoong, layering, and quote stufng which can undermine market quality
or, at the least, create a bad public perception.
As a result, minimum resting times have been suggested whereby a limit order submitted cannot be
cancelled within a given span of time. This measure can take a multitude of forms, such as a uniform
500 milli-seconds across all assets and securities, or a delay that depends on the security and/or general
market conditions. It would also be possible to prescribe different minimum resting times on limit
orders to buy or to sell, or that adjust to reect volatility or other market conditions.
6.5.2 Benefits
Minimum resting times can increase the likelihood of a viewed quote being available to trade. This
has two important benets. First, it provides the market with a better estimate of the current market
price, something which ‘ickering quotes’ caused by excessive order cancellations obfuscates. Secondly,
its visible depth at the front of the book should be more aligned with the actual depth. This knowledge
of the depth improves the ability of traders to gauge the price impact of potential trades. Quotes left
further away from the current best bid or offer are less likely to be affected by the measure since the
likelihood of them being executed within a short time is small. Nonetheless, minimum resting times
might be expected to make the order book dynamics more transparent to the market.
Minimum resting times may also reduce the excessive level of message trafc currently found in
electronic markets. Cancellations and resubmissions are a large portion of these messages, and at
peak times they can overwhelm the technological capabilities of markets (as seen for example in the
recent Facebook initial public offering (IPO) problems on NASDAQ)
26
. Some authors also suggest
that minimum resting times may reduce the protability and incidence of spoong, quote stufng and
other illicit practices. While conceptually possible, there is no clear evidence that such market abuses
only involve ickering quotes, and for those that do, surveillance and nes may prove a more efcient
deterrent than imposing a minimum resting time.
Minimum resting times may also allay concerns that markets are currently ‘unfair’ in that high frequency
traders are able to dominate trading by operating at speeds unavailable to other traders. This notion
of ‘slowing down’ markets is not generally supported by economic analyses, but it does speak to the
challenge of inducing participation if some traders, particularly small retail investors, feel that speed
makes markets unfair.
6.5.3 Costs and risks
Liquidity providers post limit orders available for trade within a period of time in return for an
expected gain in the form of the bid-ask spread. Providing limit orders is costly since posting a limit
order offers a free option to the market which is exercised at the discretion of the active trader. If an
active trader has better or newer information, the limit order poster will be adversely selected, buying
when the stock is going down and selling when the stock is going up. As with any option, its value
increases with time to maturity and with volatility. Thus, forcing a limit order to be in force longer gives
a more valuable option to the active trader, and consequently raises the cost of being a limit order
26 For discussion on the Facebook IPO, see Nanex (2012).
111
The Future of Computer Trading in Financial Markets
provider. The expected result would be an increase in the bid-offer spread or decreased depth, as
posting limit orders will be less attractive
27
.
This reluctance to post limit orders will be particularly acute during times of high volatility when the
cost of posting the option is naturally increased. This has the undesirable implication that liquidity
provision will be impeded just at the times when markets need it most. It also suggests that there could
be a feedback effect if increasing volatility triggers orders, further increasing volatility.
A minimum resting time policy may also change the dynamics of the market by attracting more
aggressive high frequency traders whose sole aim is to take advantage of the free options. Depending
on the length of compulsory resting, those limit orders close to the best bid or offer are likely to
become stale (that is, no longer at the efcient price) before they can be cancelled. This can spawn
‘front running’ by automated traders who collect the low-hanging fruit from such options. In return, the
providers of passive quotes will protect themselves against staleness through yet larger bid-ask spreads,
or by simply not posting quotes at all. Using the estimates by Farmer and Skouras
28
, the cost of hitting
such stale quotes may be as high as €1.33 billion per year in Europe
29
.
A nal argument pointing to larger spreads concerns the nature of market making in high frequency
markets. Modern market makers using HFT have, to some extent, replaced capital and inventory
capacity by speed. With minimum resting times raising the risk of losses from limit orders, high
frequency traders may reduce their market making activities and possibly be replaced by institutional
market makers, such as banks. Reduced competition among market makers and their need to earn
a return on their capital may drive up transaction costs for end users. Moreover, to the extent that
minimum resting times inhibit arbitrage between markets, which is essentially at the heart of many
HFT strategies, the efciency of price determination may be diminished.
6.5.4 Evidence
The empirical evidence to date is very limited, since there are very few natural experiments to shed
light on the costs and benets of minimum resting times
30
. In June 2009, ICAP introduced a minimum
quote lifespan (MQL) on its electronic broking services (EBS) platform. These quote requirements
set a minimum life of 250 milliseconds (ms) for their ve ‘majors’ (generally currency contracts) and
1,500ms in selected precious metals contracts. In public statements, ICAP credits the absence of a
major Flash Crash to MQLs, but of course it is difcult to know what would have happened in their
absence. To our knowledge, there has been no empirical analysis of the effects of these MQLs on
market or trader behaviour.
Until mid-2011, the Istanbul Stock Exchange (ISE) did not allow the cancellation of limit orders during
continuous auction mode unless it was the very last order entered into the system. Depending on
the structure and participation on the ISE around the switch, there may be some evidence that can
be gathered from that event. Academic research is seeking to identify the effects of this rule change,
but the results of any study would be vulnerable to criticism that they could be due to some specic
features of the ISE. Clearly, this market is rather different from the LSE or Deutsche Börse, and it is not
clear what can be learned from this experiment.
6.5.5 Conclusions
The independent academic authors who have submitted studies are unanimously doubtful that
minimum resting times would be a step in the right direction, in large part because such requirements
favour aggressive traders over passive traders and so are likely to diminish liquidity provision.
27 EIA21 describes how a minimum resting time would affect different HFT strategies such as market making, statistical arbitrage,
‘pinging’ (sending and typically immediately cancelling an order to see if hidden liquidity exists) and directional strategies. They
suggest a similar impact, although the magnitude of any effect would depend on the length of time that is imposed, market
conditions, and the liquidity of the instrument in question (Annex D refers).
28 EIA2 (Annex D refers).
29 These projected losses borne by market makers assume that market makers do not adjust their limit orders, and that
aggressive traders can wait until the end of the resting time before they hit the limit order. Competition between latency
traders would probably reduce this time, and therefore the resulting prot, by a non-negligible extent.
30 The commissioned study EIA3 (Annex D refers), examined the effects of minimum resting times inside a simulated market.
They did not recommend its adoption.
112
Economic impact assessments on policy measures
6.6 Order-to-execution ratios
6.6.1 The measure and its purpose
This measure puts an upper limit on the ratio of orders to executions, and as such is part of the larger
class of restrictions on order book activity being considered by policymakers on both sides of the
Atlantic. The idea of such restrictions is to encourage traders to cancel fewer orders, and thereby
provide a more predictable limit order book. It is hoped that such predictability will improve investor
condence in the market. As cancellations and resubmissions form the bulk of market message trafc,
this proposal would also reduce trafc and the consequent need for market participants to provide
increasing message capacity in their trading systems.
A number of exchanges have some restrictions on messages or the order-to-trade ratio (OTR). The
LSE Millennium system, for example, has message-throttling constraints which limit the total number
of messages that can come down a registered user’s pipes over a 30-second timeframe. It also has a
message pricing system which penalises excessive ordering strategies. Sponsoring rms are required
to apportion a maximum message rate threshold to prevent sponsored users from entering an
overly large number of messages. The limit is set as a maximum number of messages per second per
sponsored user and is part of the total limit allowed for the sponsoring rm’s allocation. So there are
sensible exchange-specic measures already in place that constrain the total message ow and price the
externality those messages contribute. The policy question is whether there is value in extra regulation
to enforce best practice across exchanges and extend this practice across all trading venues.
6.6.2 Benefits
Receiving, handling and storing messages is costly for exchanges, brokers and regulators. Whenever
an economic good is not priced there is a tendency to use more of it than if the user had to pay its
actual costs. If the social cost of messages exceeds its private costs, an externality results; the standard
solution is to tax messages. A ratio of orders-to-executions (OER) essentially does this, and it can
serve to align these private and social costs, thereby reducing the number of economically excessive
messages. This, in turn, will reduce the need for exchanges, brokers and other market participants to
invest in costly capacity.
With fewer quote cancellations, the order book may be less active and traders may nd it easier to
ascertain current prices and depths. An OER also may increase the likelihood of a viewed quote being
available to trade, partly because passive order submitters would focus more on those limit orders
with a higher probability of execution. An OER may also help curtail market manipulation strategies
such as quote stufng, spoong and layering. Quote stufng is when a trader sends massive numbers of
quotes and immediate cancellations, with the intention of slowing down the ability of others to access
trading opportunities. Layering refers to entering hidden orders on one side of the book (for example,
a sell) and simultaneously submitting visible orders on the other side of the book (for example, buys).
The visible buy orders are intended only to encourage others in the market to believe there is strong
price pressure on one side, thereby moving prices up. If this occurs, the hidden sell order executes,
and the trader then cancels the visible orders. Similarly, spoong involves submitting and immediately
cancelling limit orders in an attempt to lure traders to raise their own limits, again for the purpose
of trading at an articially inated price. These strategies are illegal (Trillium Trading in the USA was
recently ned by the Financial Industry Regulatory Authority (FINRA) for layering) but they are often
hard to detect. By limiting order cancellations, an order-to-execution ratio will reduce the ability to
implement these strategies.
6.6.3 Costs and risks
The nature of HFT and market making in fragmented markets naturally implies order cancellations.
Algorithmic trading, for example, seeks to reduce trade execution costs by splitting large orders into
smaller pieces and sending orders both spatially and temporally to markets. As orders execute or
languish, the execution strategy recalibrates, leading to cancellations and resubmissions. Such a trading
approach reduces execution costs for traders and leads to greater efciency in execution. Many
HFT strategies (including HFT market making) involve statistical arbitrage across markets whereby
movements in a price in one market trigger orders sent to other markets. Again, subsequent price
113
The Future of Computer Trading in Financial Markets
movements in any of the markets will trigger cancellations and resubmissions as part of the process of
reducing price discrepancies and enhancing market efciency.
Many order cancellations are a result of searching for hidden liquidity on limit order books. Exchanges
increasingly allow submitted orders to be completely hidden, meaning that the ‘best’ quotes visible on
the book are not actually the best quotes available in the market. To nd this liquidity, traders often
‘ping’ or send small orders inside the spread to see if there is hidden liquidity. Because such orders
are typically cancelled, a binding OTR would result in less pinging and, therefore, less information
extraction at the touch. As a result, more hidden orders will be posted, leading to a less transparent
limit order book. A second effect on the book may arise because orders placed away from the touch
(the best bid and ask prices) have the lowest probability of execution. In a constrained world, these
orders may not get placed, meaning that depth may be removed from the book away from the touch.
An added difculty is that the constraint may be more likely to be binding during times of extreme
market activity. Brogaard (EIA1)
31
argues that this will reduce the willingness of traders to post limit
orders during volatile times, thus reducing market liquidity provision when it is most needed.
Finally, there is the calibration exercise of where exactly to set any ratio and to what type of orders
or traders it will apply. If the upper limit of the OER is small, then it will stie legitimate activities and
prevent socially useful trading. For instance, ETFs and derivatives valuations may become unaligned,
leading to inefcient pricing. Because of this, the London Stock Exchange (LSE) has an OTR of 500/1 for
both equities, ETFs and exchange traded products (ETPs), with a high usage surcharge of ve pence for
equities and 1.25 pence for ETFs/ETPs. If instead the upper limit is set high enough not to impinge on
legitimate order strategies, it may not have much impact on the market either (a point made by Farmer
and Skourous (EIA2)
32
). If the intent is to limit manipulative strategies, a specic charge for messages
(and greater surveillance) may be a better solution
33
.
6.6.4 Evidence
There have been no published academic studies of OERs, and this greatly limits the ability to gauge the
costs and benets of order activity restrictions in general, and OERs in particular. The commissioned
study, EIA18
34
, investigates the effect of the introduction of an OER penalty regime on the Milan Borsa
on 2 April 2012. The authors’ preliminary ndings are that liquidity (spreads and depth) worsened as a
result of this policy measure. They also nd that the effect is more pronounced in large stocks, although
they acknowledge some issues with their methodology.
There are a variety of actual market programs that provide some evidence of OER impact. The ICAP
has a monthly ll ratio (MFR) requiring that at least 10% of all quotes submitted into the market must
result in an execution. Similarly, LSE’s Millennium trading system has message throttling constraints and
penalties for excessive ordering strategies. Anecdotal evidence suggests that the LSE message policy
was not fully effective in that it gave rise to new patterns of trade in low-priced stocks. The LSE has
experimented with changes in pricing effective May 4, 2010 whereby, among other measures, the
threshold for the high usage surcharge for FTSE 350 securities increased from an OTR of 100/1 to
a ratio of 500/1 (which is still the gure in use at the time of writing). The frequency of order book
updates nearly doubled for a few months as a result before coming down again. Unfortunately, we
are not aware of a proper scientic investigation of these effects.
6.6.5 Conclusions
An OER is a blunt measure that constrains both abusive and benecial strategies. It may not do too
much harm if the upper limit is large enough not to hinder market making and intermediation, but
to the extent that it is binding on those activities it may be detrimental to both spreads and liquidity.
It is unlikely that a uniform OER across markets would be optimal because it depends upon the type of
securities traded and the trader clientele in the market. If a ratio could be structured to target
31 EIA1 (Annex D refers).
32 EIA2 (Annex D refers).
33 EIA21 outlines a number of ways in which an OER could be bypassed or manipulated by HFT, whereby any benets from the
rule may be reduced (Annex D refers).
34 EIA18 (Annex D refers).
114
Economic impact assessments on policy measures
those quotes that are considered socially detrimental directly, then it might be a useful tool for
combating market manipulation. The absence of research which investigates costs and benets, as
well as the difculty of actually setting this measure optimally, suggest caution in adopting this approach
for the market.
6.7 Maker-taker pricing
6.7.1 The measure and its purpose
Maker-taker pricing refers to a fee structure in electronic markets whereby providers of liquidity via
limit order submissions (or ‘makers’) receive a rebate on their executed orders while active traders (or
‘takers’) pay a fee for executing against these limit orders. The current fee structure at the LSE (and
many other European exchanges) uses maker-taker pricing and features both a quantity discount and a
rebate for suppliers of liquidity.
The issues with maker-taker pricing are twofold. First, does this pricing scheme incentivise high
frequency traders to dominate the provision of liquidity and exclude other traders from posting limit
orders? Second, does it raises agency issues for brokers who route orders based on the net fees and
who may as a result not provide best execution for their clients?
The policy issue is whether to limit maker-taker pricing with the stated objectives of reducing the role
of HFT in markets and making order-routing decisions more transparent.
6.7.2 Benefits
In electronic markets, liquidity is provided by limit orders that are posted by passive traders willing
to provide an option to active traders. The more limit orders posted on a market, the more liquidity
there is for other traders to execute against. If there are no frictions in the market, then the bid-
ask spread would settle at a level that exactly compensates the providers of liquidity with the value
received by takers of liquidity (see Foucault (EIA12)).
Real markets do have frictions, however, so fee and pricing models are not irrelevant. By paying those
who make liquidity while charging those who take liquidity, maker-taker pricing has the potential to
improve the allocation of the economic benets of liquidity production. This in turn can incentivise
potential suppliers of liquidity and lead to faster replenishment of the limit order book. Varying maker-
taker fees with market conditions also provides a means to improve liquidity provision during times
of market stress. A recommendation to have such time-varying fee structures was one nding of the
Commodity Futures Trading Commision (CFTC)-SEC’s Task Force On Emerging Market Issues (the
Flash Crash commission).
Maker-taker pricing can also be an effective way for new market venues to compete against established
venues. Smart order-routing systems can direct order ow to venues with more aggressive pricing
models. That can in turn put pressure on fees in other markets and lead to more competitive pricing.
The success of BATS in the USA is often attributed to their aggressive use of maker-taker pricing. An
interesting competitive development in the USA has been the arrival of trading venues offering ‘taker-
maker’ pricing. In these venues, providers of active order ow get rebates while providers of passive
order ow face charges. Venues offering this pricing model are attempting to attract the ‘less toxic’
orders of retail traders, and market makers pay for the privilege of interacting with this order ow. In
US options markets, both models exist simultaneously, suggesting that a variety of pricing models may
be viable in heterogeneous markets
35
.
6.7.3 Costs and risks
High frequency traders are generally better able to put their limit orders at the top of the queue due
to their speed advantage and to their use of ‘big data’ to forecast market movements. The maker-taker
fee structure may incentivise them to do so even more, with the result that institutional investors’
limit orders are executed only if high frequency traders nd it uneconomical to do so. It follows that
institutional investors will hold back from submitting limit orders, leaving the market vulnerable to
35 Anand et al. (2011) .
115
The Future of Computer Trading in Financial Markets
transient participation by high frequency traders during times of market stress. This, in turn, could
exacerbate episodes of periodic illiquidity.
It also follows that because institutional investors will now submit more market orders, they will face
increased costs arising from bid-ask spreads, and taker fees. This problem of higher trading costs can
be compounded if the routing decisions taken by intermediaries on behalf of clients are inuenced in
a suboptimal way by the fee structure offered by disparate venues. In particular, the broker may opt
to send orders to venues offering suboptimal execution in return for rebates that are not passed on
to the originating investor. This incentive will be even greater if these rebates are volume-dependent.
Because European best execution requirements are complex, it may not be easy to monitor such
practices.
A complex system of maker-taker pricing that is context- and venue-dependent can confuse market
participants and lead to erroneous decisions. This may be particularly true if markets vary fees and
rebates across time. Because spreads can vary, it is not entirely clear how much incremental effect
on liquidity will arise from time-varying rebates.
6.7.4 Evidence
There is little evidence in the academic literature that high frequency traders are ‘abusing’ the current
fee structure. Hendershott and Riordan (2011) nd evidence that high frequency market makers lose
money in the absence of rebates
36
. This is consistent with competitive pricing strategies forcing spreads
to lower levels than they would be in the absence of maker-taker pricing.
Examining the effects of a controlled experiment on maker-taker pricing on the Toronto Stock
Exchange, Malinova and Park (2011) nd that the bid-ask spread adjusted to reect the breakdown
of maker-taker fees
37
. They also found that the quoted depth of stocks eligible for maker-taker pricing
increased signicantly, suggesting provision of greater liquidity. Adjusting for the fees, the average bid-
ask spread was the same before and after the introduction of maker-taker pricing, and volume was
greater for those stocks. Overall, maker-taker fees improve markets by increasing depth and volume
while holding spreads (including fees) the same. Anand et al. (2011) compares the make-take structure
with the traditional structure in options markets, where exchanges charge market makers and use
payments for order ow. They nd that neither structure dominates on all dimensions but that the
make-take structure is more likely to narrow existing quotes, attracts more informed order ow,
draws new liquidity suppliers, and performs better in lower tick size options.
A similar study investigated the introduction of maker-taker exchange fees for Australian securities
cross-listed on the New Zealand Stock Exchange in 2008. Berkman et al. ( 2011)
38
found that depth at
the best quotes as well as trading activity increases with the introduction of maker-taker fees, though
there is little evidence of a change in bid-ask spreads.
The only study focusing specically on maker-taker pricing in European markets identied by the
Project is Lutat (2010)
39
. He nds that the introduction of maker-taker pricing by the SWX Europe
Exchange did not affect spreads but led to an increase in the number of orders at the top of the book.
Incidentally, Menkveld (2012) nds that spreads fell dramatically when Chi-X began trading Dutch index
stocks, suggesting that its maker-taker model may have improved market competitiveness
40
.
6.7.5 Conclusion
Overall, the limited evidence suggests that maker-taker pricing improves depth and trading volume
without negatively affecting spreads.
116
36 Hendershott & Riordan (2011).
37 Malinova & Park (2011).
38 Berkman et al. ( 2011).
39 Lutat (2010).
40 Menkveld (2012).
Economic impact assessments on policy measures
6.8 Central limit order book
6.8.1 The measure and its purpose
A central limit order book (CLOB) would consolidate all limit orders into a single queue for trading.
This, in turn, would reduce the incidence of locked markets (when the best price to buy is the same
as the best price to sell) or crossed markets (when the best price to buy is lower than the best price
to sell). A central queue would also ensure that limit order providers are treated fairly in terms of price
or time priority. An actual centralised single order book is not necessary to achieve a CLOB, although
that is one option. Another alternative is a decentralised US-style consolidated quote and trade tape:
this can work to create a virtual CLOB when combined with rules specifying ‘trade at’ (the rst order
at a price must be executed rst) or ‘trade-through’ (another venue can trade an order provided they
match the better price posted on another market) protection. The policy issue is whether Europe
should establish some form of a CLOB.
6.8.2 Benefits
In Europe, best execution requirements are not formulated solely in terms of price and so linkages
between markets in Europe are not as tight as between markets in the USA. This is demonstrated by
the higher frequency of locked and crossed markets and trade-throughs in Europe, where stocks can
trade at different prices at the same time on different venues. Moving to a CLOB could reduce such
differences and improve market quality.
A CLOB also reduces queue-jumping across limit order books and therefore encourages the display
of liquidity. Having a stock trade only in a single market setting, however, represents a step backward
to the situation before the rst round of MiFID. A virtual CLOB could ease this concern, but it might
require extensive regulatory rule-making to set out how the various markets would be linked. The
growth of HFT, with the consequent increase in arbitrage across markets, may help reduce these
market abnormalities without having to establish a single CLOB. The benets of a CLOB would be
reproduced in a virtual CLOB by enhancing transparency and linkages between markets.
6.8.3 Costs and risks
Recombining local limit order books into a CLOB poses a variety of technological and economic
difculties. Creating a single pool of liquidity confers monopolistic power on a trading venue, which
allows it to increase fees and costs which are then passed on to investors. A single physical pool may
also stie competition as smaller venues may be unable to differentiate their services. MiFID was
intended, in part, to address the problems caused by the centralisation of trading.
Creating a virtual CLOB as is partially accomplished in the USA raises a host of issues. While trade-
throughs and locked and crossed markets are relatively rare in the USA, internalisation and order
preferencing
41
have ourished in part due to the access fees required by US routing rules. Moreover,
US rules provide only top of book protection, meaning that once the best posted order is executed,
the broker or venue is free to execute the rest of the order at inferior prices prevailing on their
exchange. This discourages limit order provision as better orders can be passed over if they are at
another venue.
A major problem with any centralised limit order implementation is cost. Creating a single physical
order book for all of Europe is probably unfeasible. A virtual order book created by the simultaneous
linking of all European trading venues requires both instantaneous access for all traders, a central tape
to provide information on where the best price is available and complex technology to keep track
of shifting order priorities induced by trade executions. The Project was not able to nd any cost
estimates for building such a system.
41 Order preferencing refers to arrangements whereby orders are sent to a pre-specied exchange or broker/dealer in exchange
for a per share payment. When these arrangements are made with a broker/dealer rm, the rm directly crosses these orders
with its other order ow, allowing the trade to occur internally (i.e. internalisation).
117
The Future of Computer Trading in Financial Markets
6.8.4 Evidence
There is a growing literature suggesting that current markets can achieve the consolidation benets of
a CLOB without having to move to a structure for a CLOB. Foucault and Menkveld (2008) show that
when EuroSETS introduced a competing limited order book for Dutch equities (up to then, traded only
on Euronext) spreads were largely unaffected, controlling for other factors, and consolidated market
depth was up substantially by 46% to 100%
42
. However, depth measures must be interpreted with
caution given the ephemeral nature of orders in some electronic venues. That is, the larger depth may
be posted by the same high frequency traders which may cancel quotes on one venue once the quote
on the other venue is hit.
Gresse (DR19) compares European stocks before and after the implementation of MiFID in November
2007 and shows quoted and effective bid-ask spreads are smaller on average after the implementation of
MiFID. This evidence suggests that the fragmentation of order ow has not adversely affected European
transaction costs. Degryse et al. (2011) analyse 52 Dutch stocks (large and mid-cap) over the 2006–
2009 period
43
. Moderate levels of market fragmentation (as measured by the Herndahl index) are
associated with an improvement in market liquidity, but too high a level of market fragmentation is
harmful for market liquidity, especially for the primary market. Ende and Lutat (2010), Storkenmaier and
Wagener (2011), and Gomber et al. (2012) discuss the issue of trade-throughs and locked and crossed
markets in European markets
44
. They generally nd that these market failings are not as bad as some
reports have claimed and that they have improved over time, indicating that a virtual market is emerging.
For example, in April/May 2009, 85% of quotes for FTSE100 companies coincided with a positive
spread across the main UK venues, while in April/May 2010 this percentage had increased to 94%.
6.8.5 Conclusion
The evidence suggests that the benets to consolidation of trading into a single order book have been
almost achieved by the virtual consolidation of fragmented trading venues.
6.9 Internalisation
6.9.1 The measure and its purpose
Internalisation refers to the practice whereby some customer trades are executed internally by brokers
or intermediaries and so do not reach public markets. This can provide benets to both parties of the
transaction as well as to the broker who arranges it.
Internalisation is part of a broader set of issues on order routing and pre-trade transparency. This is
because in the process of trying to nd a counterparty for a client’s order, the broker may also route
the order to dark pools that are run by groups of brokers rather than to the public limit order book.
To address this issue, MiFID introduced a category of trading venue called systematic internaliser (SI)
which deals on its own account by executing customer orders in liquid shares outside a regulated
market or a MTF, subject to certain transparency requirements. For example, Goldman Sachs and
Nomura operate SIs in the UK. Since their introduction, the volume traded on SIs in Europe has
increased a little but still represents only a small fraction of the total traded volume. In contrast, the
over-the-counter (OTC) category has grown considerably.
There are a variety of concerns connected with broker crossing networks
45
. One is that crossing
networks have avoided the burdensome rules associated with the SI category and have kept their
liquidity dark and unregulated. There are also worries that internalisation has increased to an extent
that it poses a risk to overall market quality. This is because keeping order ow out of the public limit
order book might adversely affect the liquidity of public markets and the effectiveness of the price
42 Foucault & Menkveld (2008).
43 Degryse et al. (2011) .
44 Ende & Lutat (2010) Storkenmaier & Wagener (2011), Gomber et al. (2012).
45 Broker crossing networks are trading platforms in which traders submit buy and sell orders and trades are executed for
matching orders at an exogenous price, typically the midpoint of the current bid and offer spread. As there are often more
orders on one side than the other, not all orders execute. Trades that do execute, however, do so at a better price than they
would have in the market.
118
Economic impact assessments on policy measures
discovery process. Finally, there are concerns about possible negative effects from internalisation during
a crisis period such as the Flash Crash.
The commissioned study EIA10
46
considers several specic policy measures that have been proposed
in the USA to limit or constrain internalisation. The trade-at rule would forbid internalisation unless the
venue is itself displaying the best bid or offer at the time the order is received, or provides signicant
price improvement on the order. The rationale for this rule is tied to the current US price-only best
execution system, which does not operate in Europe. The sub-penny pricing rule would allow RLPs
to quote in minimum tick sizes on public exchanges smaller than one penny, hidden from the view
of the wider market, and accessible only to providers of retail order ow. The purpose of this rule
would be to reduce the incentive to use internalised venues which can use smaller tick sizes than public
exchanges. The minimum size requirement rule would restrict the size of order that an internaliser can
execute, to direct more retail ow to the public markets. The dark pool quote threshold rule would
require internalisers to display a stock’s trading if overall trading of that stock by internalisers exceeded
a set proportion, such as 5% or 25%, in order to improve the transparency of trading.
6.9.2 Benefits
Internalisation essentially diverts order ow from public markets to private venues. This has two
undesirable effects. First, internalised orders do not contribute to the costly process of price discovery
but instead ‘free ride’ on this function performed by public markets. Exchanges and other public
markets argue that this is unfair as they are not compensated for this activity by the internalisers.
Second, to the extent that the diverted order ow comes from uninformed retail traders, it is less
‘toxic’ than order ow that is not internalised. Order ow is considered ‘toxic’ when it is related to
possible future price movements. Thus an informed trader will be buying when there is good news,
hoping to prot from the information when prices subsequently adjust. The market maker, however,
is on the other side of the trade, and so generally loses to informed traders. If order ow is too toxic,
spreads have to widen and market makers may even curtail the provision of liquidity. Exchanges argue
that internalisation, by taking the non-toxic ow, makes the orders coming to the exchanges more toxic
and reduces market quality on the exchange.
Reducing internalisation may also enhance competition in the market. In particular, price matching
preferencing arrangements eliminate the incentive to compete on the basis of price, so a reduction in
internalisation may reduce spreads. Bloomeld and O’Hara (1999) found in an experimental study that
order preferencing led to less competitive market outcomes once preferencing in the market reached a
threshold level
47
.
Restraining internalisation may reduce the potential for some conicts of interest. For example, it has
been alleged that when prices fell during the Flash Crash, internalisers executed buy orders in house
but routed sell orders to exchanges, adding further pressure on liquidity in these venues. Furthermore,
with US-style order protection rules, orders must be executed at the best bid or offer. But in times
of market stress, the consolidated tape may be slower to update than the proprietary data feeds
purchased directly from the markets. This sets up the potential for internalisers to give clients the stale
national best bid and offer price while simultaneously trading at better prices available in the market.
6.9.3 Costs and risks
In normal circumstances, internalisation is mutually benecial for participants: brokers execute client
orders, avoid paying exchange and clearing fees, and receive non-toxic trades; clients execute their
trades without signalling their intentions publicly or incurring market impact. Best execution rules in
the USA require that internalised orders are executed at the current best bid or offer. Best execution
rules in Europe differ, but internalised trades often achieve better prices and are executed inside the
spread. Thus, for many traders, internalisation is benecial. Requiring a minimum size, however, would
effectively preclude retail order ow from being internalised, while a trade-at rule would dramatically
46 EIA10 (Annex D refers).
47 Bloomeld & O’Hara (1999).
119
The Future of Computer Trading in Financial Markets
reduce the quantity of internalisation. Both changes would be expected to raise transaction costs for
retail traders.
If minimum size thresholds are mandated, institutional investors can continue to trade large blocks
of shares through internalisers, but the volume coming from retail investors will fall, leading to lower
protability for the internalisers. With a dark pool quote threshold rule, institutional investors with
a large block-trading demand may not be able to trade opaquely. This, in turn, could discourage
institutional shareholding (especially of small-cap stocks) in the rst place.
Finally, restricting internalisation would remove some liquidity providers from the market. This might
be expected to increase spreads and transactions costs for all traders.
6.9.4 Evidence
There are a variety of studies that provide evidence on this issue. Weaver (2011) analyses the effect of
the increase in internalisation on market quality using NYSE, AMEX and NASDAQ data from October
2010
48
. Controlling for variables known to be associated with spreads, he nds that internalisation (as
represented by trades reported to a trade reporting facility (or TRF)) is directly associated with wider
spreads. For example, a NYSE listed stock with 40% of TRF volume has a quoted spread $0.0128
wider on average than a similar stock with no TRF reporting. However, institutional investors trading
through internalisers may in fact have traded within the spread set on the public exchanges, so these
order book ndings cannot be used to draw conclusions about the welfare consequences as there is
no evidence given on the actual transaction costs faced by these traders. Weaver does nd that higher
levels of internalisation are always associated with higher levels of return volatility.
O’Hara and Ye (2011), however, nd very different results using TRF data
49
. These authors view TRF
reported trades as reecting fragmentation rather than internalisation per se. This is because TRF
data also include trading on non-exchange venues such as Direct Edge and BATS (both of which have
since become stock exchanges). Using a matched sample approach, O’Hara and Ye nd that stocks
with greater TRF volume had lower spreads and greater informational efciency. They interpret this as
evidence that fragmentation does not reduce market quality, but they refrain from offering conclusions
on internalisation.
A third relevant study is by Larrymore and Murphy (2009)
50
. In October 1998, the Toronto Stock
Exchange imposed a new rule that effectively banned internalisations of limit orders for 5,000 shares
or less unless they resulted in better prices. They nd a statistically signicant improvement in market
quality as a result, with spreads down, quoted depth up and volatility down.
6.9.5 Conclusion
In summary, the data relating to this measure are mixed at this point as to the impact of internalisation
on market quality. Some internalisation of trades is necessary and benecial, but there must be a limit,
which if breached would result in negative outcomes for many participants. It is harder to say where
precisely markets are currently located on this range of outcomes, so policy makers should err on the
side of caution before being over- prescriptive about these practices.
6.10 Order priority rules
6.10.1 The measure and its purpose
Order priority rules determine the sequence in which submitted orders are executed. Most exchanges
have migrated to electronic limit order books using price/time priority (PTP). This means that a limit
order is queued and executed when there is trade at that price on a rst-come rst-served basis.
However, there are exceptions or qualications to this rule in practice. Hidden orders generally
take a lower priority than visible ones, partially hidden orders are separated into the two queuing
48 Weaver (2011).
49 O’Hara & Ye (2011).
50 Larrymore & Murphy (2009).
120
Economic impact assessments on policy measures
priorities and contractual preferencing arrangements may also violate strict PTP
51
. The policy issue is
whether PTP unduly rewards high frequency traders and leads to over-investment in an unproductive
technology arms race.
6.10.2 Benefits
The greatest benet of a PTP rule is that it treats every order equally. Using other priorities, such
as a pro rata rule where every order at a price gets a partial execution, gives greater benets to large
traders over small traders. In addition, PTP provides a stronger incentive to improve the quote than a
pro rata rule, enhancing liquidity dynamics. Limit order providers face risks in that traders with better
information can prot at their expense. PTP encourages risk-taking by giving priority in execution to
limit order providers willing to improve their quotes.
However, if PTP is not operating across platforms (which is typically the case in Europe), then the
magnitude of the time priority incentive is potentially greatly reduced by a substantial increase in
fragmentation. Fragmentation reduces incentives for limit orders to compete aggressively in order
to jump the queue.
PTP is also a convenient tool used by traders for directional strategies by, for example, buying through
a market order while selling with a limit order and waiting.
6.10.3 Costs and risks
With a PTP rule, the rst order in at a price executes rst. This priority rule encourages the race for
speed, and fosters practices such as collocation with exchanges or investments in costly technology
to be earlier in the queue. There are serious concerns that the quest for relative speed leads to over-
investment in an unproductive ‘arms race’.
Alternative priority rules such as call auctions (for example, TAIFEX for trading the Taiwan Stock
Exchange Stock Index Futures) avoid the issue of the speed priority. They may reduce the race for
speed, lead to considerable aggregation of information and limited adverse selection when trading
arises, and be less prone to (mini) ash crashes. Crossing networks traditionally featured a single
clearing price (as is the case with call auctions), although contemporary crossing mechanisms feature
a variety of priority rules.
Other mechanisms have been suggested to extract information better, such as spread-priority (a
spread order is placed on both sides of the market) ahead of price and time
52
. If a person has private
information and knows that a security is underpriced, then he/she does not like to quote both ways.
Spread-priority would imply that he/she needs to put in a higher bid than he/she otherwise would
have, giving away some information. Of course, if all information is extracted, then the Grossman-
Stiglitz paradox obtains
53
.
6.10.4 Evidence
There is not much empirical research comparing the benets of different priority rules. Call auctions
are used at the open and close on many exchanges, and so investors can choose to submit orders at
those times if they prefer an alternative priority structure.
51 Preferencing arrangements generally promise to execute an order at a given price (typically the current best bid or offer at the
time of the trade). These orders violate price/time priority because they can execute before orders place rst.
52 CBOE (Chicago Board Options Exchange) Rulebook. Paragraph (e) of Rule 6.45, Priority of Bids and Offers, provides a limited
exception from the normal time and price priority rules for spread orders. Spread orders are dened in Exchange. Rule
6.53(d) to include only those spreads that involve the same class of options on both legs of the spread. Rule 24.19 provides an
exception from the normal priority rules for an OEX-SPX.
53 The Grossman-Stiglitz paradox states that informed traders will only gather new information if they can prot on its trading,
but if the stock price already incorporates all information then there is no way to prot on it. But if no one gathers new
information, how did the information get into the stock price? See driver review DR12 (Annex D refers) for a discussion of this
paradox and its role in the current high frequency environment.
121
The Future of Computer Trading in Financial Markets
6.10.5 Conclusion
Without undertaking a compelling cost-benet analysis, the benets of rule changes currently appear
insufcient relative to the costs. There does not appear to be a strong case to change substantially the
system of order priority.
6.11 Periodic call auctions
6.11.1 The measure and its purpose
Most equity trading is now undertaken through the mechanism of electronic limit order books. There
are some arguments that this incentivises expenditure designed to achieve a relative speed advantage,
and that speed of this order brings no social benet
54
. An alternative trading mechanism can be based
on periodic auctions, which can be designed to minimise the advantage of speed and to mitigate other
negative outcomes of the continuous trading model such as manipulative strategies.
The specic proposal in EIA11 involves a sequence of intraday auctions of random starting point and
duration that accumulates orders and then executes at a single price using a pro rata mechanism to
allocate outcomes. There would be a cap on the size of orders to be included and each auction would
be a sealed bid, meaning that during the bidding process no information is disseminated to bidders.
These intraday auctions would differ from existing auctions employed at the open and close of the
business day, or the auctions used to re-start trading after a trading halt.
These call auctions would replace continuous trading and serve to focus the attention of market
participants on fewer, centralised trading events. There could also be a hybrid approach in which call
auctions are interspersed throughout the day with continuous trading, but as of yet no market has
proposed or adopted such an approach.
6.11.2 Benefits
The main benet of periodic call auctions would be a reduction of the speed of trading and the
elimination of the arms race for speed discussed above. Indeed the speed of trading could be
controlled through the timing and frequency parameters which could be tuned to individual and market
conditions. It might increase liquidity or at least concentrate it at particular time points. It may also
decrease the likelihood of short but dramatic deteriorations in liquidity which have been documented
by Nanex
55
.
6.11.3 Costs and risks
While in theory a sequence of batch auctions can be Pareto optimal (improving someone’s welfare
without reducing another’s welfare), little is known about their efciency in real-world securities
markets where the structure of the economy is not common knowledge. Double auctions, on the
other hand, have been the dominant nancial market mechanism for many years, and their great
efciency in allocating resources has been well documented, leading Vernon Smith famously to describe
this nding a ‘scientic mystery’
56
.
One issue with periodic batch auctions is the increased execution risk that investors will face if they
do not know when the auctions will take place and then whether their bid will be in the winning set. If
the frequency of auctions does not reect market conditions, there may be some shortfall in matching
supply and demand.
Periodic call auctions would have a severe impact on the business model of market makers and may
reduce incentives to supply liquidity. It would also seriously affect hedgers who are trying to maintain
positions across equities and derivative markets. The current market structure allows a variety of
different trading mechanisms and allows for auction-based trading as well as the continuous limit order
book in both lit and dark modes. This proposal would limit choice and require quite a drastic change
54 See for example Haldane (2011).
55 Chapter 4 has more discussion on this. See http://www.nanex.net/ashcrash/ongoingresearch.html Accessed: 17 September 2012.
56 Smith (1982).
122
Economic impact assessments on policy measures
in the trading landscape. Furthermore, it would require coordination at the global level, which may be
difcult to achieve.
6.11.4 Evidence
Historically, auctions were more prevalent in equity markets than they are today: the Paris Bourse
relied on a daily auction until the late 1980s. These mechanisms were gradually replaced by the
electronic order book, but auction mechanisms have seen a resurgence. The NYSE, NASDAQ,
Deutsche Börse, Euronext and the LSE all have opening and closing auctions as well as continuous
trading. These exchanges now also make considerable use of auctions to restart trading after
intraday trading halts. Amihud et al. (1997) investigate the transition from daily call auction only to
daily call auction followed by continuous trading on the Tel Aviv Stock Exchange in 1987
57
. They nd
improvements in liquidity and market quality attributable to the change, which suggests that continuous
trading adds positive value.
Nowadays, auction mechanisms can vary in important directions such as the degree of transparency
reported during the bidding period and whether they have a random element to the ending time or
not. One study by Bommel and Hoffmann (2011) compares the French Euronext auctions (which
report ve levels of limit orders and have xed ending times) with the German Xetra auctions (which
only disclose the indicative price and volume and have random ending points)
58
. After controlling for
stock characteristics, the evidence shows that the Euronext auctions are more liquid, contribute more
to price discovery, and are followed by lower bid-ask spreads in the continuous trading. The more
transparent market (Euronext) is superior in terms of price and quantity discovery, and price impact,
but both auction mechanisms are followed by price reversals in continuous trading.
The only exchange that apparently makes use of a similar mechanism to the periodic call auction is
the Taiwan Stock Exchange Stock Index Futures (except that it reports indicative prices and does not
have restrictive volume caps). There does not appear to be any empirical evidence to compare current
mechanisms with the proposal discussed here. However, market participants seem to have a wide
range of choices about how and where to trade including access to auction mechanisms. Kissell and Lie
(2011), for example, document that there have been no long-term trends in auction volumes on the
NYSE or NASDAQ
59
.
6.11.5 Conclusion
The evolution of many crossing networks (see, for example, POSIT) from discrete crossing (essentially
a call market) to continuous crossing, suggests a market preference (at least among institutional traders)
against call auctions.
6.12 Key interactions
In this chapter the different measures available to policy makers have been discussed individually.
A number of key interactions between the measures are considered in the section below.
The presence or absence of circuit breakers affects almost all the other measures except perhaps
notication of algorithms. The direction of the effect is harder to determine. Having more stable and
orderly markets is likely to improve conditions for many traders, but decreasing the probability of
execution for limit orders may adversely affect particular trading strategies.
Likewise, minimum tick sizes affect almost every other measure except perhaps notication of
algorithms. The tick size affects the protability of market making and so will affect market maker
obligations. The smaller the tick size, the more onerous are the obligations to post competitive bid-
offer quotes because the return from doing so (the spread) is smaller. Internalisers and dark venues
do not currently face minimum tick size restrictions and can use smaller tick sizes to lure trades away
57 Amihud et al. (1997).
58 Bommel & Hoffmann (2011).
59 Kissell and Lie (2011).
123
The Future of Computer Trading in Financial Markets
from the public limit order book. Limiting internalisation and raising tick sizes therefore need to be
considered together.
Maker-taker pricing also becomes more powerful the larger the tick size. Therefore, encouraging
intelligent maker-taker pricing and larger minimum tick sizes need to be considered at the same time.
PTP and minimum tick sizes are also interrelated. Time priority encourages liquidity submission if the
queue cannot be jumped at low cost, and the tick is that cost. The tick size affects the protability of
market making and so will affect market maker obligations. The smaller the tick size, the more onerous
are regulations to post competitive two-sided quotes because the return from doing so (the spread)
is smaller.
Minimum resting times and minimum tick sizes may complement each other on passive orders
and possibly conict on active orders. One of the assumed benets of minimum resting times is a
slowing down of (passive) activity. Larger tick sizes have this effect as they discourage queue-jumping
and increase the value of being towards the front of the queue. Larger tick sizes make speed more
valuable as it improves the chances to be placed towards the front of the queue, but minimum resting
times make this more dangerous for the trader. In that sense, the measures are complementary since
minimum resting times blunt to some extent the speed advantage granted by larger minimum tick
sizes to faster limit order traders. But minimum resting times also make speed for market orders more
valuable as the fastest aggressive order will be rst in picking off a now stale passive order. If ticks are
larger, this opportunity will be more protable still, albeit rarer. The interaction is therefore complex
and ambiguous.
If minimum resting times and lower minimum tick sizes are introduced for a security, the benets of
either measure may not be fully reaped. The reason is that the existence of a minimum resting time
by itself tends to increase bid-offer spreads on one hand, and on the other hand as the true value
of the security moves (ceteris paribus given the minimum tick size), it is more likely to render resting
quotes stale if tick sizes become smaller. This makes ‘picking off’ more frequent (but less protable).
It follows that given minimum resting times, a reduction in minimum tick size may not lead to a
signicant reduction in spreads as passive order submitters need to protect themselves against more
frequent sniping.
OERs and larger minimum tick size both reduce trafc and complement each other. Depending on
the nonlinearities between quote volume, server speed and stability, quote stufng may become
easier given already large volumes of data, in which case a larger minimum tick size, which might be
expected to lead to less message volume, makes quote stufng more difcult. Since the OER is a blunt
instrument which may catch useful trading strategies as well, a higher minimum tick size might allow
the OERs to be larger and still accomplish its role in reducing quote stufng without inhibiting market
making too much. If it was found that minimum tick size ought to be reduced for some securities
because the spreads are articially large for liquid stocks, then an OER may help to allow a smooth
transition to lower tick sizes without an explosion of messages.
OERs and maker-taker pricing may be conicting. If the OER is close to hitting its upper bound, it may
also be possible for high frequency traders to ‘create’ executing trades if required. Such trading costs
depend on the overall fees and rebates found in maker-taker pricing. For example, if a high frequency
trader could establish that a tick inside the touch was empty, it could send both a buy and sell limit
order at that price to the venue and get credit for two transactions. Alternatively, if a high frequency
trader could predict accurately that it was at the head of the queue at the touch, it could trade with
itself (at a high probability) by sending a matching marketable order at that precise time.
Internalisation and market maker obligations also interact. Unless compensated for market maker
obligations, market makers need to make trading prots to outweigh their market making losses. But
internalisation reduces the prots from active trading. Market maker obligations and minimum resting
times clash in the sense that high frequency traders are required by market maker obligations to post
limit orders with tight spreads while minimum resting times mean that other high frequency traders
take advantage of those stale quotes. This may mean that high frequency traders snipe each other
and that in volatile markets much of the trading would comprise high frequency traders trading with
each other.
124
Economic impact assessments on policy measures
Figure 6.1: Illustrative example of key interactions between regulatory measures
Notification of
Order to
algorithms
execution ratios
Minimum
Circuit
resting
breakers
times
Minimum
tick sizes
Suggested weaker link
Suggested stronger link
Market maker
obligations
The following diagram illustrates some pairwise interactions between six regulatory measures (those
currently being considered within MiFID II). This gure is only intended to be illustrative and not
exhaustive. Also, it does not convey the detailed nature of the interactions, which will depend upon
the specic way in which the various measures would be implemented. Some of these interactions are
described in greater detail in the main text. It does, however, demonstrate that decisions concerning
the implementation of such regulatory measures should not be taken in isolation, but that such
interactions need to be carefully considered.
T
o conclude, the main lesson of these dependencies is to underscore that whatever rules are
implemented, they must be carefully calibrated against other parameters, such as the various tick sizes
and the exact circuit breaking mechanisms in the primary exchanges.
6.13 Conclusions
This chapter has considered a variety of proposals to deal with the new world of computer-based
trading in markets. In this chapter, we have summarised the views of studies directed towards
understanding the impact of these proposed changes. We further summarise our position below.
The desirability of understanding algorithmic trading strategies and their impact on the market is
laudable but achieving this through notication requirements, as, for example, currently envisioned in
MiFID II, may not be feasible given the complexity of algorithms and their interactions in the market.
Circuit breakers have a role to play in high frequency markets, and they are found in virtually all
major exchanges. Because of the interconnected nature of markets, however, there may be need for
coordination across exchanges, and this provides a mandate for regulatory involvement at least in times
of acute market stress. New types of circuit breakers triggered as problems loom rather than after they
have emerged may be particularly effective in dealing with periodic illiquidity.
Tick size policy can have a large inuence on transaction costs, market depth, and the willingness to
provide liquidity. The current approach of allowing each European trading venue to choose its own
125
The Future of Computer Trading in Financial Markets
minimum tick size has merits, but can result in unhealthy competition between venues and a race to
the bottom. A uniform policy applied across all European
60
trading venues is unlikely to be optimal, but a
coherent overall policy for minimum tick size that applies to subsets of trading venues may be desirable.
This coordinated policy could be industry-based such the one agreed to by FESE members.
The current system of exchanges determining how to structure market maker obligations and pay for
them seems to be working well for most markets. Extending those obligations more broadly across
markets and to the market making function more generally is problematic.
The aim of a more stable limit order book is laudable, and minimum resting times seem a possible
device to achieve that aim. Many of the independent academic authors have submitted studies which
are very favourable to a slowing of the markets. Nevertheless, they are unanimously doubtful that
minimum resting times would be a step in the right direction, in large part because such requirements
favour aggressive traders over passive traders and so are likely to diminish liquidity provision.
An OER is a blunt measure that catches both abusive and benecial strategies. It may not do too
much harm if the upper limit is large enough not to hinder market making and intermediation, but to
the extent that it is binding on those activities it may be detrimental to both spreads and liquidity. It
is unlikely that a uniform OER across markets would be optimal because it depends upon the type of
securities traded and the trader clientele in the market. If a ratio could be structured to target those
quotes that are considered socially detrimental directly, then it might be a useful tool for combating
market manipulation. The absence of research investigating costs and benets, as well as the difculty
of actually setting this measure optimally, suggest caution in adopting this approach for the market.
The issue of maker-taker pricing is complex and related to other issues, such as order routing and best
execution, which seems like a more promising way of constraining any negative effects of this practice.
The CLOB is of fundamental importance and everyone involved in equity trading has an interest in
nding fruitful ways of improving the performance of the virtual central limit order book that currently
operates in Europe. The debate on how best to improve shows opposing trends towards consolidation
of trading venues and fragmentation. The tension between these two trends is likely to continue in the
near future.
Internalisation of agency order ow is in principle a good thing for all parties concerned, especially
where large orders are involved. However, the trend away from pre-trade transparency cannot be
continued indenitely without detrimental effects on the public limit order book and price discovery.
Call auctions are an alternative trading mechanism which would eliminate most of the advantage for
speed in the electronic limit order book. They are widely used already in equity markets at open and
close and following a trading halt. But no major market uses them exclusively, suggesting that they do
not meet the needs of many traders. To impose call auctions as the only trading mechanism seems
unrealistic as there are serious coordination issues related to hedging strategies that would make this
undesirable.
Further to the regulatory measures outlined in this chapter, the Report needs to consider other factors;
a discussion on nancial transaction tax can be found in Box 6.1. Other considerations, such as trust in
nancial markets and standardisation are discussed in the next chapter.
126
60 This includes countries in the European Union (EU), the European Economic Area (EEA) and Switzerland.
Economic impact assessments on policy measures
Box 6.1: Financial transaction tax
On 28 September 2011 the European Commission put forward a detailed proposal for an EU-
wide nancial transaction tax
61
. The proposed tax covered a wide range of nancial instruments
and was much more comprehensive than the UK’s stamp duty on shares. One of the stated
objectives of the proposed tax was “to create appropriate disincentives for transactions that
do not enhance the efciency of nancial markets thereby complementing regulatory measures
aimed at avoiding future crises”
62
. The Commission’s targets here included short-term trading,
particularly automated and high frequency trading
63
.
Since then, it became clear that the proposal would not achieve the unanimous support of EU
Member States as required for it to be adopted across the EU
64
. A number of concerns were
raised, including: the potential negative effect on GDP, in part resulting from an increase in
the cost of capital and hence a decline in investment; the susceptibility of the tax to avoidance
through relocation unless it was adopted at a global level; and the uncertainty as to the bearers
of the economic incidence of the tax
65
. However, supporters of the proposal dismissed these
concerns
66
, and a number of Member States are currently considering implementing a nancial
transaction tax through the enhanced cooperation procedure. Subject to a number of conditions,
this procedure allows a sub-group of Member States to introduce measures that only bind the
participating states
67
.
Within this Project we were not able to second guess the details of the tax that might eventually
be adopted. These details matter since it will determine the impact of the tax.
Financial transaction taxes have been discussed, more generally, for many years. Apart from their
revenue-raising potential, the perceived corrective function of these taxes is also often cited in
their support. Proponents
68
thus argue that such taxes can produce positive effects on nancial
markets by increasing the cost and reducing the volume of short-term trading. Crucially, this
argument is based on a view of short-term trading as being mostly speculative (often supported
by or based on trading systems) and unrelated to market fundamentals. This form of trading is
thus viewed as having a negative impact on nancial markets by contributing to excessive liquidity,
excessive price volatility and asset bubbles. Furthermore, it is argued that the increasing ratio of
nancial transactions to GDP suggests considerable socially unproductive nancial activity and
hence a waste of resources. Financial transaction taxes are also seen as a way of compensating for
the fact that many nancial services are exempt from VAT.
61 European Commission (2011a).
62 Ibid. p. 2. For an assessment of the proposal’s objectives and whether a nancial transaction tax is the instrument best suited
to achieve them see Vella et al. (2011).
63 See, for example, European Commission (2011b) at pp. 27–28, 38 and 53. See also Semeta (2012). Commissioner Semeta
here explained that “a second objective [of the FTT] is to discourage unwarranted and leveraged transactions such as high
frequency trading which inate market volumes in all segments. This should complement regulatory measures and expose
nancial market actors to price signals.”
64 The opposition of some Member States, including Sweden and the UK, was evident from an early stage. See, for example,
The Guardian (2011). The matter was put beyond doubt at a meeting of the European Union Economic and Financial Affairs
Council (ECOFIN) on 22 June 2012.
65 For an outline of these concerns, as well as arguments made by supporters of the proposal, see the report by the House of
Lords, European Union Committee (2012a). See also EIA21 (Annex D refers).
66 European Union Commission (2012) See also Grifth-Jones and Persaud, (2012).
67 Article 20 of the Treaty on the European Union and articles 326 to 334 of the Treaty on the Functioning of the European Union.
68 See, for example, Schulmeister et al. (2008).
127
The Future of Computer Trading in Financial Markets
In response it has been pointed out that: not all short-term trading is ‘undesirable’; nancial
transaction taxes do not distinguish between long-term and short-term trading or between
‘desirable’ and ‘undesirable’ short-term trading; such taxes might not affect volatility or might
even increase it because they reduce liquidity in markets; asset bubbles may also develop in the
presence of high transaction costs, as documented by the recent real estate bubbles; and it is
neither obvious what the ideal ratio of nancial activity to GDP should be, nor is it clear whether
this ratio should increase or decrease over time
69
. Finally, the VAT exemption of nancial services
reects the difculties of applying VAT to margin based nancial services. Financial transaction
taxes cannot solve this problem for a number of reasons. Most importantly, nancial transactions
taxes do not tax value added in the nancial sector.
There have been a number of academic studies, both theoretical and empirical, on the effect
of nancial transaction taxes, and transaction costs more generally. These have been helpfully
reviewed by Matheson (2011) and Hemmelgarn and Nicodeme (2010)
70
. Studies nd that by
increasing transaction costs, nancial transaction taxes reduce trading volume. This generally
reduces liquidity, which in turn can slow down price discovery
71
. The theoretical literature
produces mixed results on the effect of nancial transaction taxes on volatility, with a number of
studies showing that market microstructure plays an important part
72
. The empirical literature on
the effect of transaction costs on volatility also presents contrasting results, although studies with
better data and estimation techniques seem to nd more often a positive relationship between
transaction costs and volatility. These studies mainly relate to short-term volatility
73
. Matheson
points out that there is a lack of research on the relationship between transaction costs and long-
term price volatility and price bubbles. However, transaction costs are unlikely to play a decisive
role in determining market cycles
74
. Both theoretical and empirical studies generally nd that
transaction costs, including those by transaction taxes, tend to increase the cost of capital and
lower asset prices
75
. There is also research using experimental methods to assess the effect of
transaction taxes. In a recent study
76
, the view that transactions are relocated to untaxed markets
is conrmed, as is the nding that the impact on price volatility is in general ambiguous.
Overall, these results suggest that nancial transactions taxes i) will give rise to signicant
relocation effects to tax havens and ii) cannot be expected to reduce price volatility in asset
markets
77
. Yet there are important open questions, in particular the interaction between nancial
transaction taxes, other nancial sector taxes, and regulation, and the effect of nancial sector
taxes on long term asset price movements and asset bubbles.
69 See for example, IMF (2010).
70 T. Matheson (2011); and Hemmelgarn & Nicodeme (2010).
71 See the studies reviewed in Matheson (2011) pp. 16–20. Note that some studies nd that transaction taxes can reduce or
increase liquidity depending on market microstructure: See for example, Subrahmanyam (1998).
72 See the studies reviewed in Hemmelgarn & Nicodeme (2010), pp. 34–36.
73 See Hemmelgarn & Nicodeme (2010) pp. 34–36 and the studies reviewed therein, and also the review in Matheson (2011)
pp. 20–22.
74 Matheson (2011) p. 21.
75 See the studies reviewed in Matheson (2011) pp. 14–16.
76 Hanke et al. (2010).
77 See also IMF (2010) and the assessment in Anthony et al. (2012).
128
7 Computers and complexity
Key findings
Over coming decades, the increasing use of computers and
information technology in nancial systems is likely to make
them more, rather than less complex. Such complexity
will reinforce information asymmetries and cause principal/
agent problems, which in turn will damage trust and make
the nancial systems sub-optimal. Constraining and
reducing such complexity will be a key challenge for policy
makers. Options for legislators and regulators include
requirements for trading platforms to publish information
using an accurate, high resolution, synchronised timestamp.
Improved standardisation of connectivity to trading
platforms could also be considered.
A particular issue is the vast increase in nancial data, which
are often not standardised, nor easily accessible to third
parties (for example, regulators or academics), for analysis
and research. The creation of a European nancial data
centre to collect, standardise and analyse such data should
be considered.
131
The Future of Computer Trading in Financial Markets
7 Computers and complexity
7.1 Introduction
Computers play many roles in the nancial eld. The focus, in the preceding chapters, has primarily
been on their speed which enables much faster (and usually more accurate and error free) transactions
to be done by computers than by humans. This is the basis for the computer-based trading (CBT) on
which much of this Report focuses. But information technology can provide far more than speed alone.
Computers can hold, and allow access to, far more memory than could be stored in a combination of
human brains and libraries. They can allow data not only to be accessed, but also to be manipulated
and subjected to mathematical analysis to an extent that was previously unfeasible. New professional
groups, such as quantitative analysts, or ‘quants’, in nance and various kinds of hedge funds have
rapidly developed to exploit such possibilities.
Against such advantages from computerisation, there are also some drawbacks. In particular, the
application of electronics and information technology makes nance not only more complex but also
less personal, potentially endangering the framework of trust that has provided an essential basis
to support nancial dealing. These issues are discussed in Section 7.2. Steps that could be taken to
reduce complexity and enhance efciency, through greater standardisation of computer-based nancial
language and of nancial data are discussed in Section 7.3.
7.2. Complexity, trust, and credit
There is a general and growing perception that nancial arrangements, transactions and processes
are not only complex, but have become much more so over recent decades. As Paul Volcker
(2012) wrote
1
:
Long before the 2008 nancial crisis broke, many prominent, wise and experienced
market observers used the forum provided by the now-renowned William Taylor
Memorial Lecture – which I was recently privileged to give (Volcker 2011)
2
, and around
which this article is based – to express strong concerns about the implications of the
greater complexity, the need to develop more sophisticated and effective approaches
toward risk management and the difcult challenges for supervisors
3
.
Complexity implies greater informational asymmetry between the more informed nance expert and
the less informed client. Informational asymmetry in turn causes principal/agent problems where the
informed principal is able to benet at the expense of the uninformed client.
The experience of the nancial crisis has underlined these concerns, through for example, the mis-
selling of (synthetic) collateralised debt obligations (CDOs) to unwary buyers, and the mis-selling of
certain kinds of borrowers’ insurance. As a result of these and numerous similar examples, the nancial
industry has, to some considerable extent, lost the trust of the public. That, in itself, is serious.
Analysis on trust and regulation commissioned for this Project
4
includes the claim that computerisation
and information technology reduces the need for personal interaction in nancial services, and hence
lessens the value of reputation and of longer term relationships. Several of the features of these
changing services are, at least in part, a concomitant of computerisation. Today, the withdrawal of the
personal touch, publicity about very substantial rewards to senior managers (especially after bank losses
and failures), and reports of scandals, such as attempted London Interbank Offered Rate (LIBOR)
xing, have all served to erode trust.
132
1 Volcker (2012).
2 Volcker (2011).
3 For example, Cartellieri (1996); Crockett (1998); Fischer (2002).
4 DR30 (Annex D refers).
Computers and complexity
One reason why such loss of trust matters is that almost all nancial deals involve some risks to each
of the counterparties. If the uninformed client believes that the nance professional is always trying
to abuse any trust, in order to maximise short-run prot, the client will not accept the deal so readily
in the rst place, and, if the deal is done, will be more ready to litigate on the basis of it not being
appropriate if it goes wrong. In this environment there is a danger of a ‘heads I win; tails I sue’ culture
developing, reinforced by a media and a judiciary who also doubt the morality of nance. The adverse
effect that this can have on efciency is, perhaps, best illustrated by the additional costs that litigation
imposes on US medicine. The same trend could be developing in the nance industry.
For example, the best means of saving for pensions, according to nance theory, involves a mixture of
equities and debt, with the mix depending on age. But equities are risky. So there can be a danger of
nancial advisers, in a world where trust has declined, putting clients into a portfolio consisting largely
of ‘riskless’ government bonds and bank deposits, which has no reputational risk for the adviser, but has
severe risks of ination and a poor return for the client.
The basic nancial instruments remain debt and equity. These are, in themselves, relatively easy to
understand. Debt gives a xed interest return; if this is not paid, the creditor gains certain rights of
control over the assets of the debtor. The equity holder gets the residual return, after taxes, costs
and interest. Given the uncertainty of estimating this residual return, the equity holder has underlying
governance of the enterprise, in order to constrain inefcient or self interested management.
However, given limited liability, both equity holders and managers will nd it advantageous to take on
additional risk, and debt leverage, in pursuit of a higher return on equity (RoE)
5
. This factor was a major
driver of the leverage cycle which has badly damaged the nancial systems of the developed world
over the past decade. In the absence of direct measures to restrict the incentives to take on ‘excessive’
leverage in boom periods, notably to remove the tax benet of debt relative to equity and to reduce
the built-in incentives of shareholders or managers to increase leverage in pursuit of RoE, the second
best approach has been to advocate the introduction of hybrid instruments. These have some of the
characteristics of debt, but can be like, or transform into, equity under certain (stressed) conditions.
Conditional Convertible debt instruments (CoCos) and bail-inable bonds for banks are a leading
example. Shared Appreciation Mortgages (SAMs), sometimes also called ‘Home Equity Fractional
Interest’, could be another. For government nancing, GDP bonds, where the pay-out depends on the
growth of the country’s aggregate economy also has some similarities. Assuming no rst-best reform
of the tax system and of managerial/shareholder incentives, such hybrid instruments may become more
common over the coming decade. They will also be considerably more complex, with the valuation of
CoCos, for example, depending on many details such as the trigger and conversion terms. Moreover
their introduction, could also complicate the calculation of the fundamental values of the remaining
straight equities and bonds. This extra complexity, should it occur, will add to issues about what assets it
should be appropriate for various kinds of savers to hold and for borrowers to issue.
A particular problem with SAMs is that when housing prices have risen, those that borrowed on such
a basis will claim that they were misled, misinformed, mis-sold or unable to understand, so that they
deserve to secure all the appreciation above the face value of the mortgage. The reputational and
litigation risk of such an instrument is high. Yet housing nance is in real need of reform.
The investment decision that brought down Lehman Brothers was on the future direction of the
housing market, and hence on the value of derivative instruments based on that market. Since World
War II the major nancial crises disrupting the UK economy have all been related to real estate bubbles
and crashes, with their epicentre in commercial property. These events were the Fringe Bank crisis
(1973–75), the ‘boom and bust’ (1988–1992), and the recent Great Moderation leading on to the
Great Recession (2007 onwards). During the rst part of these three periods rising property prices,
both commercial and housing, were stimulated by laxer lending conditions, driven in most cases by
competitive pressures from challenger, competitive banks aiming to transform themselves into major
nancial institutions. As the existing established oligopoly did not want to lose market share, the
procyclical shifts in lending standards became exacerbated. When the bubbles eventually burst, the
See Admati et al. (2010 and 2012).
133
5
The Future of Computer Trading in Financial Markets
challenger banks were exposed (for example, Northern Rock in 2007), while the established banks
were seriously weakened.
The subsequent fall in housing prices, combined with the difculty of raising the subsequently increased
down-payment, has typically led to a sharp decline in housing construction. The fall-off in house-
building relative to potential housing demand then creates an excess, pent-up demand for housing,
which then generates the next housing boom. On average there have been about 20 years between
successive housing crises.
The procyclicality of housing and property nance needs to be dampened. While in principle this could
be done via macro-prudential measures, the very considerable (political) popularity of the preceding
housing booms makes it unlikely that such measures would be pressed with sufcient vigour. The
unwillingness of the Financial Policy Committee (FPC) in the UK even to ask for powers to vary Loan
to Value (LTV) or Loan to Income (LTI) ratios in the current state of public opinion and acceptance is
a notable example. Alternatives may need to be sought in other forms of instrument, for example, the
Danish covered bond mechanism
6
.
For this, and other reasons, the nancial system will likely become more, rather than less, complex
in the future. Since such complexity reinforces information asymmetries, and causes principal/agent
problems which damage trust and make the nancial system sub-optimal, how can it be constrained
and reduced?
This is a large subject. It includes education and disclosure, and the need for professional bodies
to extract and to simplify the vast amounts of disclosed information that, assisted by information
technology, is now routinely required and produced within the nancial system, and which few have the
time or inclination to read, as noted in the Kay Report (2012)
7
. One of the reasons for the popularity of
Credit Rating Agencies (CRAs) was that they reduced the extensive information on credit-worthiness
to a single statistic. Although they could have produced more qualifying information there was generally
little demand from clients.
Information technology will allow the production of an even greater volume of data. While disclosure
is generally benecial to the users of nance, perhaps the greater problem from now on will be how to
interpret the masses of data that are generated. How, for example, can the regulators identify evidence
of market abuse from the vast amounts of data on orders, and transactions? Moreover, to detect
market abuse, each national regulator will need access to international market data. Otherwise the
market abuser can evade detection by transacting simultaneously in several separate national markets.
In the USA, the Ofce of Financial Research (OFR) has been commissioned by the Dodd-Frank Act
to found a nancial data centre to collect, standardise and analyse such data
8
. The increasing number
of trading platforms, scattered amongst many locations, and the vast increase in the number of data
points in each market, in large part a consequence of CBT, makes such an initiative highly desirable (see
Chapter 5, ibid). There may be a case for a similar initiative to be introduced in Europe (see Chapter 8).
Confronted with this extra complexity in nancial instruments, one proposal that is sometimes made
is that (various categories of) agents should only be allowed to hold or issue instruments which have
been approved by the authorities in advance. This contrasts with the more common position that
innovation should be allowed to ourish, but with the authorities retaining the power to ban the uses
of instruments where they consider evidence reveals undesirable effects. The former stance, however,
not only restricts innovation, but also such ofcial approval may well have unintended consequences.
6 In Denmark, mortgage banks completely eliminate market risk as the issued bonds perfectly match the loans granted. The
linking of lending and funding is what has made the Danish mortgage system unique compared with other European mortgage
systems. See https://www.nykredit.com/investorcom/ressourcer/dokumenter/pdf/NYKR_DanishCoveredBonds_WWW.pdf,
and http://www.nykredit.com/investorcom/ressourcer/dokumenter/pdf/Danish_covered_bond_web.pdf Accessed:
17 September 2012.
7 Kay (2012).
8 See OFR (2012).
134
Computers and complexity
Many of the instruments that are sometimes accused of causing nancial instability, such as credit
default swaps (CDS), securitisations of various kinds, even sub-prime mortgages, were introduced
to meet a need. Many, possibly all, of the instruments now condemned in some quarters as having
played a part in the recent global nancial crisis would at an earlier time have probably been given
ofcial approval. Ofcials have no more, probably less, skill in foreseeing how nancial instruments will
subsequently fare than CRAs or market agents.
Virtually any instrument that is designed to hedge can also be used to speculate. Since most assets are
held in uncovered long portfolios, what really undermines assets when fear takes hold is usually the
rush to hedge such uncovered long positions, rather than simple short speculation.
7.3 Standardisation
Financial complexity has increased, and will increase further. As described in the previous section
it carries with itself many potential disadvantages and risks. There are, however, standardisation
procedures that if adopted could help limit such complexity. Standards are important in both the
nancial services industry and the wider economy because they can help to engender trust, and to
ensure levels of quality. They can allow businesses and consumers to realise economies of scale, and
information to be more readily and transparently disclosed and gathered. Importantly they drive
innovation by allowing interoperability to be achieved between systems. For example, standards govern
the interoperability of the cellular networks for mobile telephone networks. Mobile telephones are
charged using a standardised micro USB charger, the use of which is dictated by European and global
standards. Standards contribute £2.5 billion in the UK’s economy
9
. As such, the UK has an active
position on standards and actively promotes their adoption and use. The need now is to do the same
in nance as in these other industries.
World War II highlighted the need for international standards, sometimes with dramatic effect. The
British Eighth Army ghting the panzers of Germany’s Afrika Korps in the North Africa campaign
was awaiting spare parts for their tanks from their American allies. However, they were found to
be unusable because of different thread pitches on the nuts and bolts used in the parts. The British
abandoned their tanks in the North African desert, as they were unrepairable
10, 11
. The American,
British and Canadian governments subsequently joined forces to establish the United Nations
Standards Coordinating Committee (UNSCC). After the war this initiative grew to include nations
from neutral and enemy countries and in 1947 became the International Organisation for Standards
(ISO) headquartered in Geneva
12
.
Standards currently play a relatively limited role in nancial services. There are some standards but
those that do exist do not cover the broad spectrum of the whole infrastructure. The same levels
of competition, innovation, transparency or the economies of scale do not appear in computer-
based trading as in many other parts of the economy. One example is FIX which is a standard used
for routing stock purchase order from investors to brokers, and reporting the execution of those
stock purchases. Anecdotal evidence suggests that FIX helped establish electronic trading and central
limit order book marketplaces. Where standards do exist in nancial services they have largely been
driven by private, rather than public benets. Whilst standards exist for order routing, they came into
existence primarily because large buy side institutions, paying substantial commissions to the broker
dealers to whom they were routing orders, required them to implement the emerging standards, or
risk seeing a reduction in their share of those commission payments.
9 Evidence base for the economic benets of standardisation, http://www.bis.gov.uk/policies/innovation/standardisation/
economic-benets
10 Robb (1956) p. 526.
11 Surowiecki (2011).
12 Buthe & Mattli (2011).
135
The Future of Computer Trading in Financial Markets
For a standard to be widely adopted it needs to satisfy two criteria; rst, it needs to be open in the
sense that it can be readily and easily adopted by all participants in the marketplace. Second, market
participants need to believe that the standard is credible (credibility here means that people believe
that the standard will be widely adopted). The credibility of a standard may be undermined by external
factors such as the Intellectual Property Rights (IPR) framework within which standards are delivered.
DR31
13
considers standards in more detail, reviews their role in the economy, their existence in
nancial services and suggests how they may play a larger role in delivering transparency and trust, and
reducing risk of nancial instability.
Standards are needed in areas of the market where they will help deliver public benets such as
greater competition, transparency, innovation and nancial stability, yet often they only arise in
those areas where their adoption will defend the economic interests of participants in the market
place. Committees, made up largely of banks and vendors, currently govern standards and there
is little representation of other stakeholders, the investors, regulators or wider society. A mandate
for standardisation initiatives which are deemed to serve the wider public interest rather than the
narrower private interest could help secure the credibility to enable widespread adoption.
More effective governance should ensure that standards address the interests of all participants
in nancial markets. Identication of areas with either high risk to nancial stability or high cost
implications for end investors is urgently required. Particular attention is needed where information
asymmetries are causing market failure and systems weakness.
Specic requirements could include:
• All trading platforms to share and publish information to an accurate, high resolution, synchronised
timestamp. This could make surveillance easier across global markets and simplify comparisons of
trading which takes place across multiple fragmented venues.
• The exploration of the potential for common gateway technology standards to ease platform
connectivity and improve the environment for collecting data required for regulatory purposes.
Clearly, common gateway technology standards for trading platforms could enable regulators and
customers to connect to multiple markets more easily, making more effective market surveillance
a possibility. They could also create a more competitive market place in surveillance solutions.
• A review of engineering processes for electronic secondary markets. Areas that create articially
high costs within the industry and areas that increase the risk of nancial instability in transaction
processing need identifying. This information could inform regulation and standardisation.
• The creation of a standard process for making order book data available to academic researchers.
This step must address the concerns of the industry on privacy but also facilitate the conduct of
academic research. This will contribute to the creation of an evidence base to help assess the need
for further action.
The initiatives of the G20 in mandating the introduction of a legal entity identier (LEI) and universal
product identier are good rst steps
14
.
7.4 Conclusions
The nancial sector over expanded in the early years of the current century, with excessive leverage
superimposed on insufcient equity. This leverage was partly enabled by complex innovations such
as securitisation, which was in turn enabled by information technology. If the incentives for excessive
leverage cannot be directly addressed, more hybrid instruments, which combine both debt and equity
characteristics, may usefully be introduced. This may be so especially in the eld of housing nance,
where shortcomings in the latest crisis have not yet been adequately addressed.
This additional complexity has reinforced a breakdown of trust which in turn has damaged the
operations of nance. A corrective step that could, and should, be taken, is to simplify (electronic)
nancial systems by the application of greater standardisation, particularly in the form of an accurate,
13 DR31 (Annex D refers).
14 Financial Stability Board (2012).
136
Computers and complexity
high resolution synchronised timestamp. Moreover, CBT, operating on many trading platforms, has
led to a vast expansion of data, which are often not standardised, nor easily accessible to third parties
(such as, regulators and academics) for analysis and research. The relevant authorities should consider
following the US example and establish a European Financial Data Centre to collect, standardise and
analyse such data.
Chapter 8 draws together the conclusions across the Report to provide policy makers with the key
ndings and future options.
137
8 Conclusions and future options
Key findings
Analysis of the available evidence has shown that computer-based trading (CBT) has several benecial
effects on markets, notably relating to liquidity, transaction costs and the efciency of market prices.
Against the background of ever greater competition between markets, it is highly desirable that any
new policies or market regulations preserve these benets. However, this Project has also highlighted
legitimate concerns that merit the close attention of policy makers, particularly relating to the possibility
of instabilities occurring in certain circumstances, and also periodic illiquidity. The following suggests
priorities for action:
A: Limiting possible future market disturbances:
• European authorities, working together, and with nancial practitioners and academics, should assess
(using evidence-based analysis) and introduce mechanisms for managing and modifying the potential
adverse side-effects of CBT.
• Regulatory constraints involving CBT in particular need to be introduced in a coordinated manner
across all markets where there are strong linkages.
• Regulatory measures for market control must also be undertaken in a systematic global fashion
to achieve in full the objectives they are directed at. A joint initiative from a European Ofce of
Financial Research and the US Ofce of Financial Research (OFR), with the involvement of other
international markets, could be one option for delivering such global coordination.
• Legislators and regulators need to encourage good practice and behaviour in the nance and
software engineering industries. This clearly involves the need to discourage behaviour in which
increasingly risky situations are regarded as acceptable, particularly when failure does not appear as
an immediate result.
• Standards should play a larger role. Legislators and regulators should consider implementing
accurate, high resolution, synchronised timestamps because these could act as a key enabling tool
for analysis of nancial markets. Clearly it could be useful to determine the extent to which common
gateway technology standards could enable regulators and customers to connect to multiple
markets more easily, making more effective market surveillance a possibility.
• In the longer term, there is a strong case for lessons to be learnt from other safety critical industries,
and to use these to inform the effective management of systemic risk in nancial systems.
B: Making surveillance of nancial markets easier:
• The development of software for automated forensic analysis of adverse/extreme market events
would provide valuable assistance for regulators engaged in surveillance of markets. This would help
to address the increasing difculty that people have in investigating events.
C: Improving understanding of the effects of CBT in both the shorter and longer term:
• Unlocking the power of the research community has the potential to play a vital role in addressing
the considerable challenge of developing better evidence-based regulation relating to CBT risks and
benets and also to market abuse. Suggested priorities include:
• developing an ‘operational process map’: this would detail the processes, systems and
interchanges between market participants through the trade life cycle, and so help to identify
areas of high systemic risk and broken or failing processes;
• making timely and detailed data across nancial markets easily available to academics, but
recognising the possible condentiality of such data.
• The above measures need to be undertaken on an integrated and coordinated international basis
to realise the greatest added value and efciency. One possible proposal would be to establish a
European Financial Data Centre.
139
The Future of Computer Trading in Financial Markets
8 Conclusions and future options
8.1 Introduction
Well functioning nancial markets are vital for the growth of economies, and the prosperity and
well-being of individuals. They can even affect the security of entire countries. But markets are
evolving rapidly in a difcult environment, characterised by converging and interacting macro- and
microeconomic forces. The development and application of new technology is arguably causing the
most rapid changes in nancial markets. In particular, high frequency trading (HFT) and algorithmic
trading (AT) in nancial markets have attracted controversy. While HFT and AT have many
proponents, some believe they may be playing an increasing role in driving instabilities in particular
markets or imposing a drag on market efciency. Others have suggested that HFT and AT may
encourage market abuse.
What is clear is that new capabilities will include the assimilation and integration of vast quantities of
data, and more sophisticated techniques for analysing news. There will be increasing availability of
substantially cheaper computing power, particularly through cloud computing. The extent to which
different markets embrace new technology will critically affect their competitiveness and therefore
their position globally: those who embrace this technology will benet from faster and more intelligent
trading systems. Emerging economies may come to threaten the long-established historical dominance
of major European and US cities.
This Report has sought to determine how computer-based trading (CBT) in nancial markets could
evolve by developing a robust understanding of its effects. Looking forward ten years, it has examined
the evidence to identify the risks and opportunities that this technology could present, notably in
terms of nancial stability
1
but also in terms of other market outcomes such as volatility, liquidity, price
efciency and price discovery, and the potential for market abuse.
In this nal chapter, the conclusions and the options for policy makers are presented in three sections.
They comprise:
• an overall assessment of whether HFT/CBT has imposed costs or benets. The challenges of
understanding the behaviour of CBT nancial markets are then considered;
• conclusions on policies that are being considered by policy makers with the goals of improving
market efciency and reducing the risks associated with nancial instability;
• consideration of how the connectivity and interactions of market participants can be mapped
and monitored so that the behaviour of CBT in nancial markets can be better understood, and
regulation better informed.
8.2 HFT: the pursuit of ever greater speed in financial transactions
In sport, the ideal is to benet from ‘taking part’; in practice, the aim is, almost always, to come rst
and win (gold). In nancial markets, the ideal is to discover the ‘fundamental’ price of assets; in practice,
the aim is, almost always, to react rst to any ‘news’ that might drive nancial prices and thereby win
gold (of a more pecuniary nature). Whether there is any social value in running some distance .001 of
a second faster, or reducing the latency of an electronic response to news by .00001 of a second, is a
moot point. But what is clear is that both efforts are driven on by basic and natural human instincts.
Trying to prevent the attempt to learn, to assess, to respond and to react to news faster than the next
agent would only drive CBT, trading volumes and liquidity to some other venue, where such assets can
be freely traded. And that market would come to dominate price setting.
140
1 A list of denitions used in this chapter can be found in Annex C.
Conclusions and future options
While the effect of CBT on market quality is controversial, the evidence suggests that CBT has
generally contributed to improvements. In particular, over the past decade, despite some very adverse
macroeconomic circumstances:
• liquidity has improved;
• transaction costs have fallen for both retail and institutional traders;
• market prices have become more efcient, consistent with the hypothesis that CBT links markets
and thereby facilitates price discovery.
In the European Union (EU) the Markets in Financial Instruments Directive (MiFID) led, as intended, to
the establishment of a wide variety of competitive trading platforms, many of these having innovative
features. Now that the same asset can be traded on many different platforms, there is a need for
arbitrage to interconnect them and HFT provides this. Indeed the growth of HFT was, in some part,
an unintended consequence of MiFID. The bulk of the evidence reviewed in Chapter 3 suggests that,
in normal circumstances, HFT enhances the quality of the operation of markets, though the returns
to HFT may come in some signicant part from buy-side investors, who are concerned that HFT may
have the effect of moving prices against them:
• Greater transparency should allow testing of this conclusion. The introduction of a synchronised
timestamp for electronic trading would help achieve this (see Section 8.4 for further discussion).
It is not so much concerns from these investors, but rather fears that, under stressed conditions, the
liquidity provided by HFT may evaporate and lead to greater volatility and more market crashes, which
provides the main driving force for new regulatory measures to manage and to modify CBT. It is clear
that there are various mechanisms whereby self-reinforcing feedback loops can develop, leading to
market crashes, and that HFT and CBT could be a factor in some of these mechanisms (see Chapter
4).
However, several caveats apply:
• Crashes occurred prior to HFT (19 October 1987 being a prime example). The idea that market
makers would, and could, previously always respond to dampen ‘fast’ markets is not supported by
analysis of past events.
• HFT is a recent phenomenon and market crashes remain rare events. There is contention about
the cause of the Flash Crash on 6 May 2010. There is as yet insufcient evidence to determine what
role HFT played either in the Flash Crash, or other mini-crashes that have occurred since HFT
became established.
• A key element of HFT, and CBT more broadly, is that they link markets together. If a (regulatory)
constraint is put on one market while another market is free of that constraint, then, especially
in stressed conditions, electronic orders and trading will migrate to the unconstrained market. If
that market is thinner, prices will move even more, and those price changes will react back to the
constrained market.
8.3 Managing and modifying CBT
There are a variety of policies being considered by policy makers with the goals of improving market
efciency and reducing the risks associated with nancial instability. This is the main subject of Chapter
6, where all the main proposals have been carefully assessed with a particular focus on their economic
costs and benets, implementation issues and empirical evidence on effectiveness.
Overall, there is general support for circuit breakers, particularly for those designed to limit
periodic illiquidity induced by temporary imbalances in limit order books. Different markets may
nd different circuit breaker policies optimal, but in times of overall market stress there is a need
for coordination of circuit breakers across markets, and this could be a mandate for regulatory
involvement. Further investigation is needed to establish how this could best be achieved in the
prevailing market structure.
141
The Future of Computer Trading in Financial Markets
There is also support for a coherent tick size policy across similar markets. Given the diversity of
trading markets in Europe, a uniform policy is unlikely to be optimal, but a coordinated policy across
competing venues may limit excessive competition and incentivise limit order provision.
There is less support for policies imposing market maker obligations and minimum resting times
on orders. The former issue runs into complications arising from the nature of high frequency market
making across markets, which differs from traditional market making within markets. Requirements to
post two-sided quotes may restrict, rather than improve, liquidity provision. Similarly, minimum resting
times, while conceptually attractive, can impinge upon hedging strategies that operate by placing orders
across markets and expose liquidity providers to increased ‘pick-off risk’ if they are unable to cancel
stale orders.
Proposed measures to require notication of algorithms or minimum order-to-execution ratios
are also not supported. The notication policy proposal is too vague, and its implementation, even if
feasible, would require excessive costs for both rms and regulators. It is also doubtful that it would
substantially reduce the risk of market instability due to errant algorithmic behaviour. An order-to-
execution ratio is a blunt instrument to reduce excessive message trafc and cancellation rates.
While it could potentially reduce undesirable manipulative trading strategies, it may also curtail
benecial strategies. There is not sufcient evidence at this point to ascertain these effects, and so
caution is warranted. Explicit fees charged by exchanges on excessive messaging and greater regulatory
surveillance geared to detect manipulative trading practices may be more desirable approaches to deal
with these problems.
Other policies remain problematic. Overall, the limited evidence suggests that maker-taker pricing
improves depth and trading volume without negatively effecting spreads. However, the issue of maker-
taker pricing is complex and related to other issues, such as order routing and best execution, which
seems like a more promising way of constraining any negative effects of this practice. The central limit
order book is of fundamental importance and everyone involved in equity trading has an interest in
changes that improve the performance of the virtual central limit order book currently operating in
Europe. The debate on how best to do so continues, in part because the opposing trends towards
consolidation of trading venues and fragmentation is likely to remain unresolved in the near future.
Internalisation of agency order ow, in principle, benets all parties involved, especially where large
orders are involved. However, the trend away from pre-trade transparency cannot be continued
indenitely without detrimental effects on the public limit order book and price discovery. Call auctions
are widely used already in equity markets at open and close and following a trading halt. But no major
market uses call auctions exclusively to trade securities. To impose them as the only trading mechanism
seems unrealistic as there are serious coordination issues related to hedging strategies that make this
undesirable.
It should be recognised that some of the above individual policy options interact with each other in
important ways. For example, the presence or absence of circuit breakers affects most other measures,
as does minimum tick sizes. Decisions on individual policies should not therefore be taken in isolation,
but should take account of such important interactions
2
.
Coordination of regulatory measures between markets is important and needs to take place at
two levels:
• Regulatory constraints involving CBT in particular need to be introduced in a coordinated
manner across all markets where there are strong linkages.
• Regulatory measures for market control must also be undertaken in a systematic global fashion
to achieve in full the objectives they are directed at. A joint initiative from a European Ofce of
Financial Research and the US Ofce of Financial Research (OFR), with the involvement of other
international markets, could be one option for delivering such global coordination.
142
2 See Chapter 6, Section 6.12 and also the supporting evidence papers which were commissioned (Annex D refers).
Conclusions and future options
8.4 Mapping and monitoring CBT
A major challenge for understanding the behaviour of today’s heavily computer-based nancial markets
is that there is no reference ‘system map’ of the network of connectivity and interactions among the
various signicant entities within the nancial markets. Consequently, there can be severe problems
in gaining access to appropriately accurate data concerning key relationships and sequences of events
in such connected markets. Volumes of market data have grown signicantly in recent years, and a
system map would need to be constantly updated to reect the dynamic development of the nancial
networks, further increasing issues of data bandwidth.
Dealing with such ‘big data’ presents signicant new challenges. These difculties are faced not only
by regulatory authorities responsible for governing the markets, but also by academics attempting to
advance our understanding of current and future market systems. Market data needs to be collected,
made available for study, analysed and acted upon in a timely fashion.
There is also a need to have a reference ‘operational process map’ of the processes that act over the
networks of the nancial system. This would set out the processes, systems and interchanges between
market participants throughout the trade life cycle. This map should aim to identify the areas of high
systemic risk and those areas with broken or failing processes. Policy makers’ attention should be directed
at the areas highlighted to determine if a policy response is appropriate, to avoid systemic failure.
Standards currently play a relatively limited role in nancial services. Certain standards are in place but
those that do exist do not cover the broad spectrum of the whole infrastructure. Particular attention
to standardisation is needed for areas where information asymmetries are causing market failure
and systems weakness (see Chapter 7). In particular, the development of accurate, high resolution,
synchronised, timestamps would make the consolidation of information from multiple sources
easier and also promote greater transparency. The former, in particular, is a priority to enable data
from diverse sources to be easily consolidated into a view that can be used as an audit trail in any
location. The use of synchronised clocks for time-stamping market event data across key European
trading venues would need to be used, so that the event-streams from European trading venues can be
brought together. This would then allow the creation of high quality European public-domain datasets
to inform policy analysis and investment strategy. Clearly it could be useful to determine the extent
to which common gateway technology standards would ease platform connectivity and improve the
environment for collecting data for regulatory purposes.
Attempting to map and monitor CBT piecemeal, country by country, or market by market, would be
wasteful and inefcient. There would be value in establishing a European Financial Data Centre (EFDC).
One purpose of this would be to ensure that, not only is sufcient data collected and kept but, just as
important, that data emanating from differing European countries are compatible and consistent.
Besides improving the collection and consistency of nancial data, an EFDC would be the natural
source whereby academics could access and utilise the data. That should be a two-way process since
academic needs and proposals should help to inform the construction and continuing evolution of the
nancial databases.
8.4.1 Monitoring CBT: analysing the data
Further research is required to develop appropriate statistical methodology for analysing high frequency,
multi-venue order book data for multiple instruments over multiple trading periods. These methods
could then be used to evaluate different policy options. They would allow accurate tracking of any
shortfalls in implementation, and the evolution of market quality, facilitate the monitoring of transaction
cost analysis, and would also increase the opportunities for ex-post detection of market abuse.
The development of automated real-time monitoring of markets is likely to be very difcult to
implement and maintain effectively. Instead, it is likely to be more cost effective to develop software
for automated forensic analysis of adverse/extreme market events, after they have occurred,
particularly if all markets share a common timestamp, and connectivity has been simplied.
Establishing timely causal explanations for such events can be very challenging for experts, even if all
143
The Future of Computer Trading in Financial Markets
relevant market data have been captured and stored for subsequent analysis. If (as may be expected)
some data are missing or corrupted, the challenges increase signicantly.
One possible exercise for an EOFR would be to build a simulator where a continuous-time limit
order book is modelled and tested with a variety of algorithms to study the systemic properties of the
nonlinear interactions between the algorithms. This would offer an opportunity for readily replicable
empirical studies. Currently, not much is known about the interactions among algorithmic CBT systems
in normal and in stressed conditions. Since the stability of markets and the condence in the markets
depends on the outcome of the markets’ continuous double auction, analyses of the impacts of various
algorithms or of various policy actions could be informed by results from simulation models. Simulators
could also be constructed to help study not only interactions among CBT algorithms, but also the
dynamics of markets populated both by CBT systems and by human traders (see DR25 and DR27).
In essence, simulation tools and techniques could enable central regulatory authorities to judge the
stability of particular nancial markets, given knowledge of the structure of those markets (i.e. the
network of interacting nancial institutions: who interacts with whom) and the processes operating
on those networks (i.e. who is doing what, and why). A family of simulation models of various types
(for example, agent-based simulations, and/or more analytical quantitative risk assessment techniques)
would need to be constructed. For researchers the challenge is how best to model and predict the
dynamics of existing markets, and how best to evaluate different management approaches and different
potential modications (for example, via running ‘what-if’ simulation models), before they are enacted
in the real world. The example of the OFR in the USA which comes under the aegis of the Financial
Stability Oversight Council (FSOC), and its data centre provides one model. The appropriate oversight
body for an EOFR would be a matter for discussion.
Finally, ‘normalisation of deviance’ was introduced in Chapter 2 and discussed further in Chapter 4
as a potential destabilising factor in nancial markets which are heavily dependent on technology that
is ‘risky’ in the technical sense that the limits of safe operating behaviour and the vulnerabilities of
the technology are not fully known in advance. Dealing with normalisation of deviance is problematic
for policy makers because it is a social issue, rooted in the culture of organisations and in failures of
communication within and between organisations. Normalisation of deviance occurs in situations
where circumstances force people to learn the technology’s limits by experience, and where
organisational culture does not place a sufciently high premium on the avoidance of adverse events
and accidents. Altering social behaviour and organisational culture via regulatory means can be difcult.
Legislators and regulators should consider how best to encourage desired accident-avoiding
practices and behaviours, and to discourage the normalisation of deviance. Safety-critical engineering
practices have been developed in aerospace, in nuclear power engineering and similar elds, in part
because of the reputational damage and legal repercussions which can follow from serious accidents.
Studies of high-reliability organisations (HROs) were discussed briey in Chapter 2 in the context of
working practices that avoid normalisation of deviance. These practices have been developed and
rened in HROs in part because of the culture and ethics of the organisations, but also because of
the fear of punitive consequences where the failure of a process can subsequently be shown to be
professionally negligent, or even criminal. Nevertheless, because normalisation of deviance occurs in
situations where people are learning the limits of a system via experience, providing incentives may
not be sufcient.
Internal working practices common in HROs and other areas of safety-critical systems engineering
are not yet widespread in the engineering teams responsible for CBT systems in nancial institutions.
Similarly, nancial regulatory authorities seem likely to benet from better understanding of regulating
safe systems in, for example, air transport or medical engineering. What is needed in the nance
industry is a greater awareness of the systemic risks introduced by normalisation of deviance, and
more dialogue with, and opportunities to learn from, other industries where safety-critical and high-
integrity engineering practices have been developed over decades. Without this, there is a possibility
that systemic risks are heightened, because there will be little or no way to predict the systemic
stability of nancial markets other than by observing their failures, however catastrophic those failures
turn out to be for individual investors or nancial institutions.
144
Conclusions and future options
Finally, computerisation has implications for the conduct of nancial markets beyond its effect on
the speed of transactions. In Chapter 7 some of these wider implications are, briey, considered,
particularly the potential role of computerisation in eroding trust in nancial markets. While this may
be the case, standards apart, it is hard to think of policies to restore trust beyond those already put
forward in the Kay Report
3
.
8.5 Conclusions
This report has presented evidence that CBT can improve the quality of markets, fostering greater
liquidity, narrowing spreads, and increasing efciency. Yet these benets may come with associated
costs: the rates at which current systems can interact autonomously with each other raises the risk
that rare but extreme adverse events can be initiated and then proceed at speeds very much faster
than humans can comfortably cope with, generating volumes of data that can require weeks of
computer-assisted analysis by teams of skilled analysts before they are understood. Although adverse
events may happen only very rarely, there is a clear danger that very serious situations can develop at
extreme speed.
This Report has suggested options for limiting possible future market disturbances, making surveillance
of nancial markets easier, and for improving the scientic understanding of how CBT systems act
and interact in both the shorter and the longer term. All of the options have been considered in the
context of a substantial body of evidence drawn from peer reviewed research studies commissioned
by the Project from a large international group of experts. The options presented here are directed
not only to policy makers and regulators but also to market practitioners.
The world’s nancial markets are engines of economic growth, enabling corporations to raise funds and
offering investors the opportunity to achieve their preferred balance of expected risks and rewards. It
is manifestly important that they remain fair and orderly. Deciding how best to ensure this, in light of
the huge growth in both the uptake and complexity of CBT that has occurred in the last decade, and
which can be expected to continue in the next, requires careful thought and discussion between all
interested parties.
Kay (2012).
145
3
Annex A Acknowledgements
1
The Government Ofce for Science would like to express its thanks to the following individuals
who were involved in the Project’s advisory bodies, undertook the detailed technical work and who
provided expert advice in the Project’s many workshops and studies. Foresight would also like to
thank the individuals who peer-reviewed individual papers, and the many others not listed here who
contributed views and advice in other ways.
HIGH-LEVEL STAKEHOLDER GROUP
Dr Greg Clark MP (Chair) Financial Secretary, Her Majesty’s Treasury
Mark Hoban MP
Ronald Arculli OBE
Dr John Bates
Prof. Sir John Beddington
CMG FRS
Dr Adrian Blundell-Wignall
Andrew Bowley
Dominique Cerutti
Stijn Claessens
Scott Cowling
Dr Claire Craig CBE
Kerim Derhalli
Richard Gorelick
Philippe Guillot
David Harding
Christopher Jackson
Christopher Marsh
John McCarthy
Robert McGrath
Nicholas Nielsen
Mark Northwood
Stephen O’Connor
Scott O’Malia
Olivier Osty
Jon Robson
Xavier Rolet
Thomas Secunda
Dr Kay Swinburne MEP
Ali Toutounchi
Danny Truell
Paul Tucker
Laurent Useldinger
Martin Wheatley
Former Financial Secretary, Her Majesty’s Treasury
and Former Chair
Chairman, Hong Kong Exchanges and Clearing
Executive Vice President and Chief Technology Ofcer,
Progress Software
Government Chief Scientic Advisor, Government Ofce
for Science
Deputy Director (Financial and Enterprise Affairs), OECD
Managing Director, Head of Electronic Trading Product
Management, Nomura
President, Deputy CEO and Head of Global Technology,
NYSE Euronext
Assistant Director, Research Department, IMF
Global Head of Scientic Equity Trading, Blackrock
Deputy Head of the Government Ofce for Science
Managing Director and Global Head of Equity Trading and
Head of Equities EMEA, Deutsche Bank
CEO, RGM Advisors LLC
Executive Director of the Markets Directorate, Autorité
des Marchés Financiers
CEO, Winton Capital
Director (Execution Sales), Merrill Lynch
Managing Director (Advanced Execution Services),
Credit Suisse
General Counsel, GETCO
Global Head of Trading, Schroders
Global Head of Trading, Marshall Wace
Global Head of Equity Trading, Fidelity
Chairman, International Swaps & Derivatives Association
Commissioner, Commodity Futures Trading Commission
Deputy Head of Global Equities & Commodity
Derivatives, BNP Paribas
President of Enterprise, Thomson Reuters
CEO, London Stock Exchange
Founding Partner, Bloomberg
European Parliament
Managing Director (Index Funds), Legal & General
Investment Management
Chief Investment Ofcer, Wellcome Trust
Deputy Governor, Bank of England
CEO, ULLink
Managing Director, Consumer and Markets Business Unit, FSA
Individuals are only listed once. Where an individual contributed in multiple ways (for example, producing an evidence paper
and attending a workshop) they are only listed for their rst recorded contribution.
147
1
The Future of Computer Trading in Financial Markets
LEAD EXPERT GROUP
2
Dame Clara Furse (Chair)
Professor Philip Bond
Professor Dave Cliff
Professor Charles Goodhart
CBE, FBA
Kevin Houstoun
Professor Oliver Linton FBA
Dr Jean-Pierre Zigrand
Non-executive Director, Legal & General plc, Amadeus
IT Holding SA, Nomura Holdings Inc., Chairman, Nomura
Bank International, Non-executive Director, Department
for Work and Pensions and Senior Adviser, Chatham House.
Visiting Professor of Engineering, Mathematics and
Computer Science at the University of Bristol and Visiting
Fellow at the Oxford Centre for Industrial and Applied
Mathematics.
Professor of Computer Science at the University of Bristol.
Professor (Emeritus) of Banking and Finance at the
London School of Economics.
Chairman of Rapid Addition and co-Chair, Global
Technical Committee, FIX Protocol Limited.
Chair of Political Economy at the University of Cambridge.
Reader in Finance at the London School of Economics.
Foresight would like to thank Dr Sylvain Friederich, University of Bristol, Professor Maureen O’Hara,
Cornell University and Professor Richard Payne, Cass Business School, City University, London for their
involvement in drafting parts of this Report.
Foresight would also like to thank Andy Haldane, Executive Director for Financial Stability at the Bank
of England, for his contribution in the early stages of the Project.
AUTHORS AND CONTRIBUTORS TO THE EVIDENCE BASE – DRIVER REVIEWS
Professor Khurshid Ahmad
Professor Michael Aitken
Dr Ali Akyol
Professor Peter Allen
Professor James Angel
Angelo Aspris
Professor Robert Battalio
Professor Alessandro Beber
Professor Itzhak Ben-David
Dr Daniel Beunza
Professor Bruno Biais
Professor Robin Bloomeld
Professor Ekkehart Boehmer
Dr Jonathan Brogaard
Professor Seth Bullock
Trinity College, Dublin
University of New South Wales
The University of Melbourne
Craneld University
Georgetown University
The University of Sydney Business School
Finance and Presidential Faculty Fellow at the University
of Notre Dame
City University, London
The Ohio State University
The London School of Economics
The Toulouse School of Economics
City University, London
Ecole des Hautes Etudes Commerciales du Nord
University of Washington
School of Electronics and Computer Science, University
of Southampton
2 The following acknowledges associations of the Lead Expert Group members:
• Dame Clara Furse is a non-executive Director of Nomura Holdings, Legal & General Group, Amadeus IT Holdings and the
Department for Work and Pensions. Dame Clara is also a Senior Advisor to Chatham House.
• Professor Phillip Bond has not declared any association that has bearing on their involvement with the Project.
• Professor Dave Cliff has not declared any association that has bearing on their involvement with the Project.
• Professor Charles Goodhart is a part-time Senior Economist Consultant to an investment bank in London and an occasional
economic adviser to a couple of nancial institutions in London, all of which may, and some of which do, use CBT techniques.
• Kevin Houstoun is a part-time Chairman of Rapid Addition Limited, http://www.rapidaddition.com/, a leading vendor of
messaging components to the trading industry. Many of Rapid Additions clients use CBT techniques. Kevin also consults
to investment banks, stock exchanges, vendors and the buy side on messaging and strategy; aspects of this work include
CBT. Kevin has an minority investment in Thematic Networks, http://www.thematic.net/, the leading supplier of bespoke
enterprise social network platforms. Thematic networks has investments in many trade publications including HFT Review,
http://www.hftreview.com/, and Money Science, http://www.moneyscience.com/. Kevin was a partner in Fourth Creek, a FX
high frequency trading rm, prior to its sale to ICAP in 2008. Kevin is actively involved in standardisation within the trading
space and current co-Chairs the FIX Protocol Limited Global Technical Committee, he has held this position since 2004.
• Professor Oliver Linton has not declared any association that has bearing on their involvement with the Project.
• Dr Jean-Pierre Zigrand has not declared any association that has bearing on their involvement with the Project.
148
Annex A Acknowledgments
Calder OBE, FRSE University of Glasgow
Cartlidge University of Bristol
Comerton-Forde Australian National University
Professor Muffy
Dr John
Professor Carole
Professor Gregory
Professor Rama
Professor Douglas
Dr Jon
Marco
Professor Doyne
Professor Karl
Sean
Kingsley
Professor Thierry
Dr Sylvain
Professor Alex
Gérard
Professor Michael
Professor Carole
Professor Antonio
Dr Greg
Professor Frederick
Professor Terrence
Professor David
Professor Giulia
Kevin
Professor Nicholas
Professor Neil
Professor Charles
Dr Guray
Torben
Professor Hayne
Professor Marc
Professor Andrew
Professor Terry
Professor Katya
Professor Konstantinos
Professor Peter
Professor John
Professor Thomas
Professor Antonio
Professor Albert
Dr Yuval
Professor Gautam
Dr Richard
Dr Juan Pablo
Professor Andreas
Professor David
Professor Christine
Professor Richard
Anders
Dr Steve
Dr Alex
Connor
Cont
Cumming
Danielsson
de Luca
Farmer
Felixson
Foley
Fong
Foucault
Friederich
Frino
Gennotte
Goldstein
Gresse
Guarino
Gyurko
H.deB. Harris
Hendershott
Hutchison
Iori
James
Jennings
Johnson
Jones
Kucukkocaoglu
Latza
Leland
Lenglet
Lepone
Lyons
Malinova
Manolarakis
McBurney
McDermid
McInish
Mele
Menkveld
Millo
Mitra
Olsen
Pardo-Guerra
Park
Parkes
Parlour
Payne
Pelli
Phelps
Preda
National University of Ireland
Columbia University
York University
London School of Economics
Budget Planning and Control Analyst at Alenia
North America
Santa Fe Institute
Hanken School of Economics
Capital Markets Cooperative Research Centre
Australian School of Business
Ecole des Hautes Études Commerciales de Paris
University of Bristol
University of Sydney
Ecole d’Economie de Paris
Babson College
Université Paris-Dauphine
University College, London
University of Oxford
Wake Forest University
University of California, Berkeley
Lancaster University
City University, London
Southampton University
Miami University
Columbia University
Baskent University
University of Bristol
University of California, Berkeley
Head of Department (Management and Strategy)
at European Business School
University of Sydney
University of Oxford
University of Toronto
Visiting researcher at Imperial College, London;
MANOLOTECH LTD
King’s College, London
Leader of the High Integrity Systems Engineering Group,
University of York.
University of Memphis
University of Lugano and Swiss Finance Institute
VU University of Amsterdam
The London School of Economics
Brunel University; CEO OptiRisk Systems
Chairman and Chief Executive Ofcer, Olsen Ltd
The London School of Economics
University of Toronto
Harvard University
University of California, Berkeley
Cass Business School, City University, London
University of Essex
University of Edinburgh
149
The Future of Computer Trading in Financial Markets
Professor Tom Rodden University of Nottingham
Professor Bernd Rosenow University of Leipzig
Professor Leonard Rosenthal Bentley University
Professor Gideon Saar Cornell University
Professor Patrik Sandås University of Virginia
Professor Alexander Schied University of Mannheim
Dr Kevin Sheppard University of Oxford
Professor Hyun Song Shin Princeton University
Professor Spyros Skouras Athens University
Professor Didier Sornette Chair of Entrepreneurial Risks, ETH Zurich
Professor Chester Spatt Carnegie Mellon University
Dr Sasha Stoikov Head of Research, Cornell Financial Engineering
Manhattan
Professor Martyn Thomas CBE Martyn Thomas Associates Ltd
Professor Ian Tonks University of Bath
Professor Edward Tsang University of Essex
Dr Dimitrios Tsomocos University of Oxford
Nick Vanston Experian Decision Analytics
Professor Daniel Weaver Rutgers, Ccpd/Technical Training
Professor Ingrid Werner Ohio State University
Dr Alexander Wissner-Gross Institute Fellow at the Harvard University Institute
for Applied Computational Science
Julie Wu
Professor Mao Ye University of Illinois
Carla Ysusi Financial Services Authority
Ilknur Zer London School of Economics
Feng Zhan
Professor Frank Zhang Yale University
Guannan Zhao University of Miami
AUTHORS AND CONTRIBUTORS TO THE EVIDENCE BASE – IMPACT ASSESSMENTS
Professor Lucy Ackert Kennesaw State University
Professor Amber Anand Syracuse University
Professor Marco Avellaneda New York University
Professor Jean-Philippe Bouchaud Chairman of Capital Fund Management; Professor,
Ecole Polytechnique
Dr Paul Brewer Economic and Financial Technology Consulting
Professor Charles Cao Pennsylvania State University
Professor Tarun Chordia Emory University
Professor Kee Chung Department Chair, Finance and Managerial Economics,
University of Buffalo
Professor Hans Degryse Tilburg University
Professor Lawrence Glosten Columbia University
Anton Golub Manchester Business School
Professor Roger Huang University of Notre Dame
Professor Sebastian Jaimungal University of Toronto; President and CEO, Tyrico
Financial Services Inc.
Dr Elvis Jarnecic University of Sydney
Professor Kenneth Kavajecz University of Wisconsin
Professor John Keane University of Manchester
Professor Yong Kim University of Cincinnati
Professor Paul Kofman University of Melbourne
Professor Bruce Lehmann University of California, San Diego
Dr Wai-Man Liu Australian National University
Dr Marco Lutat University of Frankfurt
Professor Rosario Mantegna University of Palermo
150
Annex A Acknowledgments
Patrice Muller Partner, London Economics Ltd
Professor Gulnur Muradoglu City University, London
Professor Marco Pagano University of Naples Federico II
Professor Marios Panayides University of Pittsburgh
Professor Ser-Huang Poon The University of Manchester
Professor Ryan Riordan Karlsruhe Institute of Technology
Professor Ailsa Roell Columbia University
Professor Andriy Shkilko Wilfrid Laurier University
Professor Avanidhar Subrahmanyam University of California, Los Angeles
Professor Erik Theissen University of Mannheim
John Willman Editorial Consultant
Professor Jimmy Yang Oregon State University
CO-AUTHORS AND CONTRIBUTORS TO THE EVIDENCE BASE
Professor Ashok Banerjee Financial Research and Trading Lab, IIM Calcutta
Professor Peter Bossaerts California Institute of Technology
Dr Dan Brown Entrepreneur in Residence, University College London
Professor Jaksa Cvitanic California Institute of Technology
Dan di Bartolomeo President and founder, Northeld Information Services
Professor Clemens Fuest Oxford University Centre for Business Taxation
Dr Stefan Hunt Financial Services Authority
Dr Paul Irvine University of Georgia
Professor Donald MacKenzie University of Edinburgh
Professor Linda Northrop Carnegie-Mellon University
Professor Maureen O’Hara Cornell University
Professor Aaron Pitluck Illinois State University
Professor Charles Plott California Institute of Technology
Professor Andy Puckett University of Tennessee
Charlotte Szostek University of Bristol
Professor Philip Treleaven University College London
John Vella Oxford University Centre for Business Taxation
Professor Kumar Venkataraman Southern Methodist University
Susan von der Becke ETH Zurich
Dr Anne Wetherilt Senior Manager, Payments and Infrastructure Division,
Bank of England
Xiang Yu Brunel University
AUTHORS AND CONTRIBUTORS TO THE EVIDENCE BASE – ATTENDED LEG MEETINGS
Martin Abbott CEO, London Metals Exchange
Fod Barnes Senior Adviser, Oxera
David Bodanis PA Consulting
Rodrigo Buenaventura Head of Markets Participants Division, ESMA
Dr Alvaro Cartea Oxera
Edward Corcoran Her Majesty’s Treasury
Dr Luis Correia da Silva Oxera
Dr Vince Darley Head of Analytics and Optimisation, Ocado
Laurent Degabriel Head of Investment and Reporting Division,ESMA
Simon Friend Financial Services Authority
Dr Martina Garcia Her Majesty’s Treasury
Bob Giffords Independent analyst and consultant for European banking
Alasdair Haynes Chi-X Europe Limited
Nichola Hunter Head of EBS product management, ICAP
Dr Andrei Kirilenko Commodity Futures Trading Commission
Shrey Kohli London Stock Exchange
David Lawton Financial Services Authority
Dr Charles-Albert Lehalle Head of Quant Research, Cheuvreux Credit Agricole
151
The Future of Computer Trading in Financial Markets
Remco Lenterman EPTA
Roger Liddell Member, CFTC Global Markets Advisory Committee
Graham Lloyd PA Consulting
Ted Moynihan Oliver Wyman
John Romeo Oliver Wyman
Tim Rowe Financial Services Authority
John Schoen Global Head of EBS Spot Platform and Market Risk, ICAP
Tom Stenhouse London Stock Exchange
Kee-Meng Tan Knight Equity Markets Europe
Steven Travers Head of Regulatory Law and Strategy,
London Stock Exchange
Samad Uddin International Organization of Securities Commissions
Reinder Van Dijk Oxera
Ken Yeadon Managing Partner and Co-Founder,
Thematic Capital Partners
AUTHORS AND CONTRIBUTORS TO THE EVIDENCE BASE – WORKSHOPS:
SCOPING WORKSHOP
Professor David Arrowsmith Queen Mary College, University of London
Richard Balarkas CEO, Instinet Group Incorporated
Dr Robert Barnes UBS
Graeme Brister Legal First
Dr Mario Campolargo European Commission
Dr Christopher Clack University College London
Robert Crane Goldman Sachs
Rod Garratt University of California, Santa Barbara
Dr Kerr Hatrick Deutsche Bank
Rebecca Hodgson Facilitation Team
Professor Julian Hunt FRS Global Systems Dynamics
Karl Massey Deutsche Bank
Professor Philip Molyneux Bangor University
Paul Ormerod Volterra Consulting
Bill Sharpe Facilitation Team
Dr Christopher Sier Technology Strategy Board, Knowledge Transfer Network
Michael Straughan Financial Services Authority
Stuart Theakston GLC Ltd
Natan Tiefenbrun Turquoise
Nigel Walker Technology Strategy Board, Knowledge Transfer Network
WORKSHOP ON DRIVERS OF CHANGE
Dr Abhisek Banerjee The London School of Economics
Dr Christian Brownlees Universitat Pompeu Fabra
Svetlana Bryzgalova The London School of Economics
Ziad Daoud The London School of Economics
Professor Frank De Jong Tilburg University
Eric Hunsader Founder and CEO, Nanex
Bonsoo Koo The London School of Economics
Professor Albert Kyle University of Maryland
Professor Otto Loistl Vienna University of Economics and Business
Katrin Migliorati University of Bergamo
Dr Sujin Park The London School of Economics
Philippe Vandenbroeck Consultant
Claire Ward Financial Services Authority
Yang Yan The London School of Economics
152
Annex A Acknowledgments
WORKSHOP ON COMPLEX SYSTEMS
Dr Alfonso Dufour University of Reading
Dr Graham Fletcher Craneld University
Susan Gay Financial Services Authority
Dr Enrico Gerding University of Southampton
Professor Iain Main University of Edinburgh
Professor Sheri Markose University of Essex
Professor Robert May University of Oxford, Workshop Speaker
Edgar Peters First Quadrant
Chris Sims Ignis Asset Management
Dr Ryan Woodard ETH, Zurich, Workshop Speaker
INDUSTRY WORKSHOP, LONDON
Professor Neil Allan Systemic Consult Ltd
Richard Anthony HSBC
Dr Valerie Bannert-Thurner European Operations, FTEN
Matthew Beddall Winton Capital
Aileen Daly Morgan Stanley Europe
Adrian Farnham Turquoise MTF
Dr Rami Habib Algo Engineering
Peter Hafez RavenPack Spain
Mark Harding Head of EMEA Equity Dealing, Invesco
Lee Hodgkinson SVP, Head of European
Relationship Management, NYSE Euronext
Ashok Krishnan Bank of America Merrill Lynch
Miles Kumaresan Algonetix UK
Mike Lovatt Accenture
Dan Lyons RGM Advisors
Dr Mridul Mehta Citadel Execution Services Europe
Paul Milsom Legal & General plc
Alasdair Moore Business Development Director and Founder, Fixnetix
Andrew Morgan Deutsche Bank
Landis Olson Hudson River Trading
Fred Pedersen Business Development Manager, Vincorex
Mike Powell Thomson Reuters
Peter Randall Equiduct Europe
Dr Riccardo Rebonato RBS Global Banking & Markets
Antoine Shagoury London Stock Exchange
David Shrimpton Goldman Sachs
Murray Steel Man AHL
Dr Alex Suesserott RTS Realtime Systems
David Surkov Egham Capital Europe
Adam Toms Co-Global Head of Electronic Sales, Nomura Europe
Laetitia Visconti Barclays Capital
INDUSTRY WORKSHOP, SINGAPORE
Robert Bilsland Tullett Prebon
Tim Cartledge Barclays Capital
Tze Ching Chang Capital Markets Department
Emma Chen Prime Minister’s Ofce
Eng Yeow Cheu A*STAR
Sutat Chew Singapore Exchange
Wei Min Chin Accenture
Bill Chow Hong Kong Exchanges and Clearing Limited
Michael Coleman Aisling Analytics
153
The Future of Computer Trading in Financial Markets
Petar Dimitrov
Jerome Guerville
Masayuki Kato
Eugene Khaimson
Abhishek Mohla
Jonathan Ng
Richard Norris
Ivan Nosha
Julian Ragless
Sunarto Rahardja
Dr Sintiani Dewi Teddy
Peter Tierney
Yeong Wai Cheong
Rebecca Weinrauch
INDUSTRY WORKSHOP, NEW YORK
Paul Adcock
Irene Aldridge
Lloyd Altmann
Scott Atwell
Joe Bracco
Thomas Chippas
Sanjoy Choudhury
Dr Ian Domowitz
Joseph Gawronski
Bryan Harkins
Peter Lankford
James Leman
Deborah Mittelman
Craig Mohan
Chris Morton
Adam Nunes
Florencia Panizza
John Prewett
Adam Sussman
Greg Tusar
WORKSHOP OF CHIEF ECONOMISTS
Dr Angus Armstrong
Dr Riccardo Barbieri
Malcolm Barr
Nick Bate
Roger Bootle
George Buckley
Willem Buiter
Robert Carnell
Dr Jonathan Cave
James Derbyshire
Peter Dixon
Patrick Foley
Robert Gardner
Jonathan Gershlick
Dr Larry Hatheway
Dr Andrew Hilton OBE
Dr Haizhou Huang
RSR Capital
BNPP
Tokyo Stock Exchange
Deutsche Bank
UBS
Prime Minister’s Ofce
Nomura
NOVATTE
Hong Kong Exchanges and Clearing Limited
Chi-East
A*STAR
NYSE Euronext
Algorithms Trading and Research
GETCO
NYSE Euronext
Able Alpha Trading
Accenture
American Century Investments
BATS Trading
Barclays Capital
Nomura
Investment Technology Group
Rosenblatt Securities
Direct Edge
Securities Technology Analysis Center
Westwater Corp.
UBS
CME Group Inc.
Thomson Reuters
Hudson River Trading
MarketAxess
Citigroup
TABB Group
Goldman Sachs
Director of Macroeconomic Research,
National Institute of Economic and Social Research
Chief European economist, Mizuho International
Economist, JPMorgan Chase
Economist, Bank of America Merrill Lynch
Economist, Capital Economics
Chief UK Economist, Deutsche Bank
Chief Economist, Citigroup
Chief International Economist, ING
Senior Research Fellow, RAND (Europe)
Senior Analyst, RAND (Europe)
Global Equities Economist, Commerzbank AG
Chief Economist, Lloyds Banking Group
Chief Economist, Nationwide Building Society
Department for Business, Innovation and Skills
Chief Economist, UBS
Director, Centre for Study of Financial Innovation
Chief Strategist, China International Capital Corporation
154
Annex A Acknowledgments
Dr Brinda Jagirdar
Dr Stephen King
Dr Eduardo Loyo
Dr Gerard Lyons
Dr Andrew McLaughlin
Professor David Miles
Barry Naisbitt
Dr Jim O’Neill
Mario Pisani
Professor Vicky Pryce
Dr Richard Roberts
Des Supple
David Tinsley
Yan Wang
Dr Martin Weale
Dr Ohid Yaqub
General Manager and Head of Economic Research,
State Bank of India
Group Chief Economist, HSBC
Chief Economist, BTG Pactual
Chief Economist, Standard Chartered
Chief Economist, The Royal Bank of Scotland Group
Monetary Policy Committee, Bank of England
Chief Economist, Santander UK
Chief Economist, Goldman Sachs
Her Majesty’s Treasury
Senior Managing Director, FTI Consulting
SME Economist, Barclays
Head of Fixed Income Research, Nomura International
Chief UK Economist, BNP Paribas
Bank of China
Monetary Policy Committee, Bank of England
RAND Europe
ADDITIONAL AUTHORS AND CONTRIBUTORS TO THE EVIDENCE BASE – WORKING
PAPER (not mentioned elsewhere)
Professor Michael Kearns University of Pennsylvania
Professor Fritz Henglein University of Copenhagen
ADDITIONAL CONTRIBUTORS – GOVERNMENT OFFICE FOR SCIENCE
Michael Reilly Foresight Research and Knowledge Manager
Alun Rhydderch Foresight International Futures Manager
Karen Wood Foresight Communications Manager
Martin Glasspool Foresight Central Team
Dr Jane Jackson Foresight Central Team
The Government Ofce for Science would also like to thank the 104 anonymous interviewees who
participated in a survey of end users undertaken by Oliver Wyman
3
; the anonymous computer-based
traders who presented their views during interviews conducted by Beunza, Millo and Pardo-Guerra
4
;
and the academics and industry practitioners who contributed to studies by London Economics
5
,
and Oxera
6
.
3 SR1 (Annex D refers).
4 IN1 (Annex D refers).
5 EIA22 (Annex D refers).
6 EIA21 (Annex D refers).
155
The Future of Computer Trading in Financial Markets
Annex B References
Abad, D. and Pascual, R. (2007) Switching to a temporary call auction in times of high uncertainty. Available:
http://www.cnmv.es/DocPortal/Publicaciones/MONOGRAFIAS/Mon2007_19en.pdf
Accessed: 22 August 2012.
Abolaa, M. (2001) Making markets: opportunism and restraint on Wall Street. Cambridge: Harvard
University Press.
Admati, A.R, DeMarzo P.M., Hellwig, M.F. and Peiderer, P. (2010) Fallacies, irrelevant facts,
and myths in the discussion of capital regulation: why bank equity is not expensive. Available:
https://gsbapps.stanford.edu/researchpapers/library/RP2065R1&86.pdf
Accessed: 07 September 2012.
Admati, A.R, DeMarzo P. M., Hellwig, M.F. and Peiderer, P. (2012) Debt overhang and capital
regulation. Available:
http://www.gsb.stanford.edu/news/packages/PDF/AdmatiDebt032612.pdf
Accessed: 07 September 2012.
AFM (Netherlands Authority for Financial Markets), (2010) High frequency trading: the application
of advanced trading technology in the European marketplace. Available:
http://www.afm.nl/~/media/les/rapport/2010/hft-report-engels.ashx
Accessed: 22 August 2012.
Amihud, Y., Mendelsohn, H. and Lauterbach, B. (1997) Market microstructures and securities values:
evidence from the Tel Aviv Stock Exchange. Journal of Financial Economics, 45(3): 365–390.
Anand, A., McCormick, T. and Serban, L. (2011) Does the make-take structure dominate the traditional
structure? Evidence from the options markets (27 June 2011) Social Science Research Network.
Available:
http://ssrn.com/abstract=2074318
Accessed: 22 August 2012.
Angel, J. (1994) Tick size, share prices, and stock splits. Working Paper, Georgetown University. Journal
of Financial Abstracts, 1(30). Available:
http://bubl.ac.uk/archive/journals/jfa/v01n3094.htm
Social Science Research Network.
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=5486
Accessed: 16 August 2012.
Angel, J., Harris, L. and Spatt, C. (2010) Equity trading in the 21st century. Social Science Research
Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1584026 http://dx.doi.org/10.2139/ssrn.1584026
http://www.sec.gov/comments/s7-02-10/s70210-54.pdf http://1.usa.gov/LZJf10)
Accessed: 16 August 2012.
Antony, J., Bijlsma, M., Elbourne, A., Lever, M. and Zwart, G. (2012) Financial transaction tax: review and
assessment. CPB Discussion Paper 202, CPB Netherlands Bureau for Economic Policy Analysis. Available:
http://www.cpb.nl/sites/default/les/publicaties/download/discussion-paper-202-nancial-transaction-
tax-review-and-assessment.pdf
Accessed: 10 September 2012.
Arnuk, S. and Saluzzi, J. (2012) Broken Markets: How High Frequency Trading and Predatory Practices
on Wall Street are Destroying Investor Condence and Your Portfolio. Financial Times Press.
BATS Trading Limited. (2009) Pan European Tick Size Pilot, An analysis of results. Available:
http://www.batstrading.co.uk/resources/participant_resources/BATSEuro_Tick_Size_Paper.pdf
Accessed: 22 August 2012.
Berkman, H., Boyle, G. and Frino, A. (2011) Maker-taker exchange fees and market liquidity: evidence
from a natural experiment. Queenstown, New Zealand: 2011 Financial Management Association
International (FMA) Asian conference, (6–8 April 2011). Available:
http://www.fma.org/NewZealand/Papers/GlennBoylePaper.pdf
Accessed: 17 September 2012.
156
Annex B References
Biais, B., Foucault, T. and Moinas, S. (2011) Equilibrium high frequency trading. Available:
http://www.wsuc3m.com/biais%20foucault%20moinas%2014-sept-2011.pdf
https://studies2.hec.fr/jahia/webdav/site/hec/shared/sites/foucault/acces_anonyme/Document/
Working%20papers/biais_foucault_moinas_20_oct_2011.pdf
Accessed: 22 August 2012.
Bloomeld, R. and O’Hara, M. (1999) Market transparency: who wins and who loses? Review of Financial
Studies 12(1): 5–35 DOI:10.1093/rfs/12.1.5.
Boehmer, E., Fong, K. and Wu, J. (2012) International evidence on algorithmic trading: Edhec Working
Paper. Social Science Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2022034
Accessed: 22 August 2012.
Bommel, J. van. and Hoffmann, P. (2011) Transparency and ending times of call auctions: a comparison
of Euronext and Xetra. LSF Research Working Paper Series No. 11-09.
Boston Consulting Group (2011) US Securities & Exchange Commission, Organizational Study and
Reform. Boston Consulting Group report, 10 March 2011. Available:
http://www.sec.gov/news/studies/2011/967study.pdf
Accessed: 23 August 2012.
Brogaard, J. (2010) High frequency trading and its impact on market quality. Social Science Research
Network. Available:
http://ssrn.com/abstract=1641387
http://www.fsa.gov.uk/static/FsaWeb/Shared/Documents/pubs/consumer-research/jonathan-
brogaard-hft.pdf
Accessed: 23 August 2012.
Brogaard, J., Hendershott, T. and Riordan R. (2012) High frequency trading and price discovery. Social
Science Research Network. Available:
http://faculty.haas.berkeley.edu/hender/HFT-PD.pdf
http://ssrn.com/abstract=1928510 or http://dx.doi.org/10.2139/ssrn.1928510
Accessed: 23 August 2012)
Brunnermeier, M. and Pedersen, L. (2009) Market liquidity and funding liquidity. Review of Financial
Studies, 22(6): 2201–2238.DOI:10.1093/rfs/hhn098
Buthe, T. and Mattli, W. (2011) The new global rulers, the privatization of regulation in the world
economy. Princeton University Press, ISBN 978-0-691-14479-5.
Cartellieri, U. and Greenspan, A. (1996) Global risk management. William Taylor Lecture Series ISBN:
1-56708-098-7.
Carlsson, H. and Van Damme, E. (1993) Global games and equilibrium selection. Econometrica, 61(5):
989–1018.
Castura, J., Litzenberger, R. Gorelick, R. and Dwivedi, Y. (2010) Market efciency and microstructure
evolution in U.S. equity markets: a high-frequency perspective. Working paper, RGM Advisors, LLC.
Available:
http://www.rgmadvisors.com/docs/MarketEfciencyStudyOct2010.pdf
Accessed: 16 August 2012
Cave, T. (2010) Is the party ending for high-frequency traders? Financial News 18 October 2010.
Available:
http://www.enancialnews.com/story/2010-10-18/high-frequency-party-over
Accessed: 16 August 2012
Chaboud, A., Chiquoine, B., Hjalmarsson, E. and Vega, C. (2009) Rise of the machines: algorithmic
trading in the foreign exchange market. International Finance Discussion Paper No. 980, Board of
Governors of the Federal Reserve System. http://www.federalreserve.gov/pubs/ifdp/2009/980/ifdp980.
pdf
Social Science Research Network. Available:
http://ssrn.com/abstract=1501135
Accessed: 16 August 2012.
Chicago Board Options Exchange Rulebook (CBOE). Available:
http://cchwallstreet.com/CBOE/Rules
Accessed: 16 August 2012.
157
The Future of Computer Trading in Financial Markets
Cliff, D. (2009) ZIP60: Further explorations in the evolutionary design of trader agents and online
auction–market mechanisms. IEEE Transactions on Evolutionary Computation, 13(1): 3–18.
Comerton-Forde, C., Hendershott, T., Jones, C.M., Moulton, P.C. and Seasholes, M.S. (2010) Time
variation in liquidity: the role of market maker inventories and revenues. Journal of Finance, 65(1):
295–331.
Crockett, A. (1998) Banking supervision and nancial stability. William Taylor Memorial Lecture. Available:
http://www.bis.org/speeches/sp981022.htm
Accessed: 22 August 2012.
Danielsson, J. and Shin, H.S. (2003) Endogenous risk, modern risk management: a history. London: Risk Books.
Danielsson, J., Shin, H.S., and Zigrand, J.-P. (2004) The impact of risk regulation on price dynamics,
Journal of Banking & Finance, 28(5): 1069–1087.
Danielsson, J., Shin, H.S. and Zigrand, J.-P. (2011) Balance sheet capacity and endogenous risk. Mimeo
(August 2011). Available:
http://www.riskresearch.org/les/JD-HS-JZ-39.pdf http://www.princeton.edu/~hsshin/www/
balancesheetcapacity.pdf
Accessed: 22 August 2012.
Das, R., Hanson, J., Kephart, J. and Tesauro G. (2001) Agent-human interactions in the continuous
double auction, Proceedings of IJCAI-01, (2): 1169–1176. Available:
http://ijcai.org/Past%20Proceedings/IJCAI-2001/PDF/IJCAI-2001-m.pdf/IJCAI-2001-m.pdf
http://www.research.ibm.com/infoecon/paps/AgentHuman.pdf
Accessed: 22 August 2012.
Degryse, H.A., Jong, F.C.J.M. de and Kervel, V.L. van. (2011) The impact of dark and visible
fragmentation on market quality. Social Science Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1816434
Accessed: 10 September 2012.
Dufe, D. (2010) Presidential address: Asset price dynamics with slow-moving capital. Journal of Finance,
65(4): 1237–1267. DOI: 10.3905/jot.2011.6.2.008.
Easley, D., Lopez de Prado, M. and O’Hara, M. (2011a) The microstructure of the ‘Flash Crash’:
ow toxicity, liquidity crashes and the probability of informed trading. The Journal of Portfolio
Management, 37(2): 118–128.
Easley, D., Lopez de Prado, M. and O’Hara, M. (2011b) The exchange of ow toxicity. Journal of Trading,
6 (2): 8-13.
Economist (The). (2012) Fine and Punishment. The Economist (21 July 2012). Available:
http://www.economist.com/node/21559315
Accessed: 22 August 2012.
Ende, B. and Lutat, M. (2010) Trade-throughs in European cross-traded equities after transaction
costs-empirical evidence for the EURO STOXX 50. 2nd International Conference: The Industrial
Organisation of Securities Markets: Competition, Liquidity and Network Externalities; Frankfurt.
Available:
https://www.econstor.eu/dspace/bitstream/10419/43270/1/63950793X.pdf
Accessed: 10 September 2012.
European Commission (2004). Directive 2004/39/EC of the European Parliament and of the Council of
21 April 2004 on markets in nancial instruments. Available: http://eur-lex.europa.eu/LexUriServ/
LexUriServ.do?uri=CELEX:32004L0039:EN:HTML
Accessed: 25 September 2012
European Commission (2010). Public Consultation, Review of the Markets in Financial Instruments
Directive (MiFID). Available:
http://ec.europa.eu/clima/consultations/0013/consultation_paper_en.pdf
Accessed: 27 September 2012.
European Commission (2011a) Staff Working Paper Impact Assessment (SEC(2011) 1217 nal)
accompanying the documents: Proposal for a Regulation of the European Parliament and of the Council
on Insider Dealing and Market Manipulation (market abuse) and Proposal for a Directive of the European
Parliament and of the Council on criminal sanctions for Insider Dealing and Market Manipulation p 15.
Available:
http://ec.europa.eu/internal_market/securities/docs/abuse/SEC_2011_1217_en.pdf
Accessed: 22 August 2012.
158
Annex B References
European Commission. (2011b) Staff Working Paper, Impact Assessment (SEC(2011) 1226 nal)
accompanying the documents. Proposal for a Directive of the European Parliament and of the Council
on Markets in Financial Instruments [Recast] and Proposal for a Regulation of the European Parliament
and of the Council on Markets in Financial Instruments. Available:
http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=SEC:2011:1226:FIN:EN:PDF
Accessed: 22 August 2012.
European Commission Taxation and Customs Union. (2012) Taxation of the nancial sector. Available:
http://ec.europa.eu/taxation_customs/taxation/other_taxes/nancial_sector/index_en.htm.
Accessed: 10 September 2012.
European Securities and Markets Authority (2011). Consultation Paper (Reference ESMA/2011/224).
Guidelines on systems and controls in a highly automated trading environment for trading
platforms, investment rms and competent authorities. Available:
http://www.esma.europa.eu/system/les/2011-224.pdf
Accessed: 1 October 2012.
European Union, The Council of the, Economic and Financial Affairs (ECOFIN). (2012) Press Release
3178th Council meeting (22 June 2012). Available:
http://www.consilium.europa.eu/uedocs/cms_data/docs/pressdata/en/econ/131141.pdf
Accessed: 10 September 2012.
Financial Stability Board. (2012) A Global Legal Entity Identier for Financial Markets. Available:
http://www.nancialstabilityboard.rg/publications/r_120608.pdf
Accessed: 25 September 2012
Finansinspektionen (Swedish Government). (2012) Investigation into high frequency and algorithmic
trading. Available:
http://www..se/upload/90_English/20_Publications/10_Reports/2012/htf_eng.pdf
Accessed: 22 August 2012.
Fischer, S. (2002) Basel II: Risk management and Implications for banking in emerging market countries.
William Taylor Memorial Lecture. Available:
http://www.iie.com/Fischer/pdf/scher091902.pdf
Accessed: 22 August 2012.
Foucault, T. and Menkveld, A.J. (2008) Competition for order ow and smart order routing systems.
Journal of Finance, 63(1): 119–158.
Guardian, The. (2011) Osborne wins wide support in ght against Tobin tax. Available:
http://www.guardian.co.uk/business/2011/nov/08/tobin-tax-osborne-eu-clash
Accessed: 10 September 2012.
Gennotte, G. and Leland, H. (1990) Market liquidity, hedging, and crashes. American Economic Review,
80(5): 999–1021.
Gomber, P., Arndt, B., Lutat, M. and Uhle, T. (2011) High frequency trading. Technical Report, Goethe
Universität & Deutsche Börse, Social Science Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1858626
Accessed: 22 August 2012.
Gomber, P., Pujol, G. and Wranik, A. (2012) Best execution implementation and broker policies in
fragmented European equity markets. International Review of Business Research Papers, 8(2):
144–162.
Grifth-Jones, S. and Persaud, A. Financial Transaction Taxes (2012). Available:
http://policydialogue.org/les/publications/FINANCIAL_TRANSACTION_TAXES_Grifth_Jones_
and_Persaud_February_2012_REVISED.pdf
Accessed: 10 September 2012.
Grossklags, J. and Schmidt, C. (2006) Software agents and market (in)efciency: a human trader experiment.
IEEE Transactions on Systems, Man and Cybernetics, Part C (Applications and Reviews), 36(1): 56–67.
Grossman, S. (1976) On the efciency of competitive stock markets where trades have diverse
information, Journal of Finance, 31(2): 573–585.
Grossman, S. and Stiglitz, J. (1980) On the impossibility of informationally efcient markets. American
Economic Review, 70(3): 393–408.
159
The Future of Computer Trading in Financial Markets
Haldane, A.G. (2009) Rethinking the nancial network. Available:
http://www.bankofengland.co.uk/publications/speeches/2009/speech386.pdf
Accessed: 22 August 2012.
Haldane, A.G. (2011) The race to zero. Available:
http://www.bankofengland.co.uk/publications/speeches/2011/speech509.pdf
Accessed: 22 August 2012.
Haldane, A.G. and May, R.M. (2011) Systemic risk in banking ecosystems. Nature, 469(7330): 351–355.
Halpern, J. and Moses, Y. (1990) Knowledge and common knowledge in a distributed environment.
Communications of the ACM, 37(3): 549–587.
Hanke, M., Huber, J., Kirchler, M., and Sutter, M. (2010) The economic consequences of a Tobin Tax –
An experimental analysis. Journal of Economic Behaviour and Organisation, 74 (1): 58-71.
Harris, L.E. (1997) Order exposure and parasitic traders. Available:
http://www-rct.usc.edu/~/harris/ACROBAT/Exposure.pdf
Accessed: 25 September 2012.
Hasbrouck, J. (2010) Trading costs: a rst look. Available:
http://pages.stern.nyu.edu/~jhasbrou/Teaching/2010%20Winter%20Markets/PDFHandouts/
Trading%20costs.pdf
Accessed: 15 August 2012.
Hasbrouck, J. and Saar, G. (2011) Low-latency trading, Working paper. Cornell University, Social Science
Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1695460
Accessed: 15 August 2012.
Hemmelgarn, T. and Nicodeme, G. (2010) The 2008 Financial Crisis and Taxation Policy. EU
Commission Taxation Papers, Working Paper No. 20, 2010. Available:
http://ec.europa.eu/taxation_customs/resources/documents/taxation/gen_info/economic_analysis/
tax_papers/taxation_paper_20_en.pdf
Accessed: 10 September 2012.
Hendershott, T., Jones, C. and Menkveld, A. (2011) Does algorithmic trading improve liquidity. Journal of
Finance, 66(1): 1–33.
Hendershott, T. and Menkveld, A. (2011) Price pressures. Social Science Research Network. http://ssrn.
co m /a bs t r a c t=14119 43
Accessed: 15 August 2012.
Hendershott, T. and Riordan, R. (2011) Does algorithmic trading and the market for liquidity. Journal of
Financial and Quantitative Analysis (Forthcoming article). Available:
http://faculty.haas.berkeley.edu/hender/ATMonitor.pdf
Accessed: 10 September 2012.
Hirshleifer, J. (1971) The private and social value of information and the reward to incentive activity.
American Economic Review, 61, 561–574.
Holley, E. (2012) More research needed to prevent HFT abuse – Nomura. The Trade. Available:
http://www.thetradenews.com/newsarticle.aspx?id=8645&terms=nomura
Accessed: 18 September 2012.
House of Lords European Union Committee. (2012a) 29th Report of Session 2010–12 Towards a
Financial Transaction Tax? Available:
http://www.publications.parliament.uk/pa/ld201012/ldselect/ldeucom/287/287.pdf
Accessed: 10 September 2012.
House of Lords European Union Committee. (2012b) 2nd Report of Session 2012-13 MiFID II:
Getting it Right for the City and EU Financial Services Industry. Available:
http://www.publications.parliament.uk/pa/ld201213/ldselect/ldeucom/28/28.pdf
Accessed: 25 September 2012.
Hunsader, E. (2010) Analysis of the ‘Flash Crash’ Date of Event: 20100506, Complete Text. Nanex Corp.
Available:
http://www.nanex.net/20100506/FlashCrashAnalysis_CompleteText.html.
Accessed: 18 September 2012.
160
Annex B References
International Monetary Fund (IMF) (2010) A fair and substantial contribution by the nancial sector
Final Report for the G-20 (June 2010). Available:
http://www.imf.org/external/np/seminars/eng/2010/paris/pdf/090110.pdf
Accessed: 10 September 2012.
International Organization of Securities Commission (2011) FR09/11, Regulatory Issues Raised by
the Impact of Technological Changes on Market Integrity and Efciency. Report of the Technical
Committee. Available:
http://www.iosco.org/library/pubdocs/pdf/IOSCOPD361.pdf
Accessed: 18 September 2012.
ITG (Investment Technology Group). (2011) Trading Patterns, Liquidity, and the Citigroup Split.
Available:
http://www.itg.com/news_events/papers/CitiSplit2.pdf.
Accessed: 18 September 2012.
Jones, C.M. (2002) A century of stock market liquidity and trading costs. Working paper, Social Science
Research Network. Available:
http://ssrn.com/abstract=313681 http://dx.doi.org/10.2139/ssrn.313681
Accessed: 18 September 2012.
Jopson, B. (2012) Amazon ‘robo-pricing’ sparks fears. Financial Times (8 July 2012). Available:
http://www.ft.com/cms/s/0/26c5bb7a-c12f-11e1-8179-00144feabdc0.html#axzz24IAnRMrV.
Accessed: 18 September 2012.
Joshua, C. and Stafford, E. (2007) Asset re sales (and purchases) in equity markets. Journal of Financial
Economics, 86 (2): 479–512.
Jovanovic, B. and Menkveld, A. (2011) Middlemen in limit-order markets. Social Science Research
Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1624329.
Accessed: 18 September 2012.
Kaminska, I. (2009) HFT in Europe. Financial Times Alphaville Blog. Available:
http://ftalphaville.ft.com/blog/2009/07/24/63651/high-frequency-trading-in-europe/.
Accessed: 18 September 2012.
Kaminska, I. (2011) Algo trading and the Nymex. Financial Times Alphaville Blog. Available:
http://ftalphaville.ft.com/blog/2011/03/04/505021/algo-trading-and-the-nymex/.
Accessed: 18 September 2012.
Kay, J. (2012) Interim review of UK equity markets. Available:
http://www.bis.gov.uk/assets/biscore/business-law/docs/k/12-631-kay-review-of-equity-markets-
interim-report.pdf.
Accessed: 18 September 2012.
Kay, J. (2012) The Kay Review of UK Equity Markets and Long-Term Decision Making. Final Report (July
2012). Available:
http://www.bis.gov.uk/assets/biscore/business-law/docs/k/12-917-kay-review-of-equity-markets-nal-
report.pdf.
Accessed: 17 September 2012.
Kearns, M., Kulesza, A. and Nevmyvaka, Y. (2010) Empirical limitations of high frequency trading
protability. The Journal of Trading, 5(4): 50–62. DOI: 10.3905/jot.2010.5.4.050. Available:
http://www.cis.upenn.edu/~mkearns/papers/hft_arxiv.pdf.
Accessed: 18 September 2012.
Khandani, A. and Lo, A. (2007) What happened to the quants in August 2007? Journal of Investment
Management, 5(4): 29–78.
Khandani, A. and Lo, A. (2011) What happened to the quants in August 2007? Evidence from factors
and transactions data. Journal of Financial Markets, 14(1): 1–46.
Kim, K.A. and Park, J. (2010) Why do price limits exist in stock markets? A manipulation-based
explanation. European Financial Management, 16(2): 296–318. Available:
h t t p: //d x .d o i. or g /10 .1111/ j.14 68 - 03 6X . 2 0 0 8 . 0 0 45 6 . x .
Accessed: 18 September 2012.
Kirilenko, A., Kyle, A., Samadi, M. and Tuzun, T. (2011) The Flash Crash: the impact of high frequency
trading on an electronic market. Social Science Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1686004.
Accessed: 18 September 2012.
161
The Future of Computer Trading in Financial Markets
Kissell, R. and Lie, H. (2011) U.S. exchange auction trends: recent opening and closing auction behavior,
and the implications on order management strategies. Journal of Trading, 6(1): 10–30 DOI: 10.3905/
jot.2011.6.1.010
Krugman, P. (2009) Rewarding bad actors. New York Times (2 August 2009). Available:
http://www.nytimes.com/2009/08/03/opinion/03krugman.html?_r=1.
Accessed: 18 September 2012.
Kyle, A. P. (1985) Continuous auctions and insider trading. Econometrica, 53: 1315 –1335.
Larrymore, N.L. and Murphy, A.J. (2009) Internalization and market quality: an empirical investigation.
The Journal of Financial Research, 32(3): 337–363.
DOI : 10 .1111/j .1475 - 68 03 . 20 0 9. 012 53 . x .
Lee, C., Mucklow, B. and Ready, M. (1993) Spreads, depths, and the impact of earnings information: an
intraday analysis. Review of Financial Studies, 6(2): 345–374.
DOI:10.1093/rfs/6.2.345.
Leinweber, D.J. (2009) Nerds on Wall Street: Math, Machines and Wired Markets. John Wiley Publishers.
London Stock Exchange Group plc (2011) Response to ESMA Consultation: Guidelines on systems
and controls in a highly automated trading environment for trading platforms, investment rms
and competent authorities ESMA/2011/224, http://www.londonstockexchange.com/about-the-
exchange/regulatory/lsegresponsetoesmaconsultationonsystemsandcontrols.pdf
Accessed: 10 September 2012.
Lutat, M. (2010) The Effect of Maker-Taker Pricing on Market Liquidity in Electronic Trading Systems –
Empirical Evidence from European Equity Trading. Social Science Research Network. Available:
http://ssrn.com/abstract=1752843.
Accessed: 18 September 2012.
Madhavan, A. (2011) Exchange-Traded Funds, Market Structure and the Flash Crash. Working Paper.
BlackRock, Inc. Social Science Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1932925.
Accessed: 18 September 2012.
Malinova, K. and Park, A. (2011) Subsidizing Liquidity: The Impact of Make/Take Fees on Market Quality.
Social Science Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1787110.
Accessed: 10 September 2012.
Matheson, T. (2011) Taxing Financial Transactions: Issues and Evidence. IMF WP/11/54 (IMF, March 2011).
Available:
http://www.imf.org/external/pubs/ft/wp/2011/wp1154.pdf.
Accessed: 10 September 2012.
Mehta, N. High Frequency Firms Tripled Trades Amid Rout, Wedbush Says. Bloomberg (12 August
2011). Available:
http://www.bloomberg.com/news/2011-08-11/high-frequency-rms-tripled-trading-as-s-p-500-
plunged-13-wedbush-says.html.
Accessed: 18 September 2012.
Menkveld, A. J. (2012). High Frequency Trading and the New-Market Makers, Working Paper. Social
Science Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1722924.
Accessed: 18 September 2012.
Milgrom, P. and Stokey, N. (1982) Information, trade and common knowledge. Journal of Economic
Theory, 26(1):17–27.
Mill, John Stuart (1846) A system of logic: ratiocinative and inductive; being a connected view of the
principles of evidence and the methods of scientic investigation. Harper & Brothers.
Morris, S. and Shin, H. (2002) Global games: theory and applications. In Dewatripont, L. and Turnovsky,
S. (eds). Advances in Economics and Econometrics, the Eighth World Congress. Cambridge University
Press.
Nanex. (2012) Facebook IPO How HFT Caused the Opening Delay, and Later Beneted at the Retail
Customer’s Expense. Available:
http://www.nanex.net/aqck/3099.html.
Accessed: 18 September 2012.
162
Annex B References
O’Hara, M. (1995) Market Microstructure Theory. Blackwell Publishers.
O’Hara, M. and Ye, M. (2011) Is market fragmentation harming market quality? Journal of Financial
Economics, 100(3): 459–474. Social Science Research Network. Available:
http://ssrn.com/abstract=1356839.
Accessed: 10 September 2012.
Ofce of Financial Research. (2012) Annual Report 2012. Available:
http://www.treasury.gov/initiatives/wsr/ofr/Documents/OFR_Annual_Report_071912_Final.pdf.
Accessed: 17 September 2012.
Oxera. (2011) What would be the economic impact of the proposed nancial transaction tax on the
EU? (December 2011). Available:
http://www.oxera.com/Publications/Reports/2012/What-would-be-the-economic-impact-of-the-
proposed-.aspx.
Accessed: 17 September 2012.
Oxera. (2012) What would be the economic impact on the EU of the proposed nancial transaction
tax? (June 2012). Available:
http://marketing.oxera.com/Eventsbr.nsf/0/bf32bb01f3bb843280257a23005a1e50/$FILE/Oxera%20
review%20of%20latest%20FTT%20commentary.pdf.
Accessed: 10 September 2012.
Oxford English Dictionary, Available:
http://www.oed.com/view/Entry/203876?rskey=MxYIeu&result=1&isAdvanced=false#eid b.
Accessed: 25 September 2012.
Perez, E. (2011) The Speed Traders. McGraw-Hill Publishers.
Perrow, C. (1984) Normal Accidents: Living with High-Risk Technologies. New York. Basic Books.
Robb, H. W. (1956) Signicance of Company and National Standards to Industrial Management. Harper
& Brothers.
Rothschild, M. and Stiglitz, J.E. (1970) Increasing risk: I. A denition. Journal of Economic Theory, 2(3):
225–243.
Rubinstein, A. (1989) The electronic mail game: Strategic behavior under almost common knowledge.
American Economic Review, 79(3): 385–391.
Samuelson, P.A. (1947) Foundations of economic analysis. Harvard University Press.
Schulmeister, S., Schratzenstaller M. and Picek, O. A (2008) General Financial Transaction Tax. Motives,
Revenues, Feasibility and Effects. Research Study WIFO. (March 2008). Available:
http://www.wifo.ac.at/wwa/jsp/index.jsp?typeid=8&display_mode=2&d=23923&id=31819.
Accessed: 10 September 2012.
Semeta, A. (2012) Financial Transactions Tax: The Way Ahead. Meeting of Members of the Finance and
Fiscal Committees in the Danish Parliament, Copenhagen, (19 March 2012). Available:
http://europa.eu/rapid/pressReleasesAction.do?reference=SPEECH/12/196.
Accessed: 18 September 2012.
Shin, H.S. (1996) Comparing the robustness of trading systems to higher order uncertainty. Review of
Economic Studies, 63(1): 39–59.
Shin, H.S. (2010) Risk and liquidity. 2008 Clarendon Lectures in Finance, Oxford University Press.
Sloan, S. (2012) High-frequency trading has ‘disconnected’ public, NYSE CEO says. Bloomberg (20 June
2012). Available:
http://www.bloomberg.com/news/2012-06-20/high-frequency-trading-has-disconnected-public-
nyse-ceo-says.html.
Accessed: 18 September 2012.
Smith, V.L. (1982) Microeconomic systems as an experimental science. American Economic Review, 72(5):
923–955.
Storkenmaier, A. and Wagener, M. (2011) Do we need a European ‘National Market System’?
Competition, arbitrage, and suboptimal executions. Social Science Research Network. Available:
http://www.efmaefm.org/0EFMAMEETINGS/EFMA%20ANNUAL%20MEETINGS/2011-Braga/
papers/0069_update.pdf.
Accessed: 10 September 2012.
163
The Future of Computer Trading in Financial Markets
Subrahmanyam, A. (1998) Transaction Taxes and Financial Market Equilibrium. Journal of Business, 71(1):
81–118. Available:
http://www.jstor.org/stable/10.1086/209737.
Accessed: 11 September 2012.
Surowiecki, J. (2011) Turn of the century (April 2011). Available:
http://www.wired.com/wired/archive/10.01/standards_pr.html.
Accessed: 18 September 2012.
TABB Group. (2012) US equities market 2012: mid-year review. Available:
http://www.tabbgroup.com/PublicationDetail.aspx?PublicationID=1129&MenuID=44&ParentMenuI
D=2&PageID=43.
Accessed: 18 September 2012.
Thurner, S., Doyne Farmer, J. and Geanakoplos, J. (2012) Leverage causes fat tails and clustered
volatility. Quantitative Finance, 12(5): 695–707 DOI:10.1080/14697688.2012.674301.
Troianoivski, A. (2012) Networks built on milliseconds. The Wall Street Journal (30 May 2012). Available:
http://on.wsj.com/JUMhiH.
Accessed: 16 August 2012.
US Commodity Futures Trading Commission and the US Securities & Exchange Commission. (2010)
Findings regarding the market events of May 6, 2010. Available:
http://www.sec.gov/news/studies/2010/marketevents-report.pdf
Accessed: 24 September 2012.
US Commodity Futures Trading Commission. (2010) Glossary. Available:
http://www.cftc.gov/ucm/groups/public/@educationcenter/documents/le/cftcglossary.pdf
Accessed: 1 October 2012.
US Securities & Exchange Commission. (2010) Concept release on equity market structure. Release No.
34-61458; File No. S7-02-10, page 45.
US Securities & Exchange Commission. (2012a) SEC approved the NYSE proposed pilot on July 3,
2012. Release No. 34-67347; File Nos. SR-NYSE-2011-55: SR-NYSE Amex-2011-84
US Securities & Exchange Commission. (2012b) Self-Regulatory Organizations; The NASDAQ Stock
Market LLC; NYSE Arca, Inc. Order Instituting Proceedings to Determine Whether to Approve or
Disapprove Proposed Rule Changes Relating to Market Maker Incentive Programs for Certain Exchange-
Traded Products Release No. 34-67411; File Nos. SR-NASDAQ-2012-043: SR-NYSE Arca-2012-37).
Available:
http://www.sec.gov/rules/sro/nasdaq/2012/34-67411.pdf.
Accessed: 18 September 2012.
Vaughan, D. (1997) The Challenger launch decision: risky technology, culture and deviance at NASA.
University of Chicago Press.
Vella, J., Clemens, F. and Schmidt-Eisenlohr. T. (2011) The EU Commission’s Proposal for a Financial
Transaction Tax. British Tax Review 2011 (6): 607-621
Volcker, P.A. (2011) Three years later: unnished business in nancial reform. William Taylor Memorial
Lecture. Available:
http://graphics8.nytimes.com/packages/pdf/business/23gret.pdf and
http://www.group30.org/images/PDF/ReportPDFs/GRP30_WTML13_Volcker_FNLlo.pdf.
Accessed: 18 September 2012.
Volcker, P.A. (2012) Unnished business in nancial reform. International Finance, 15(1): 125–135. DOI:
10 .1111/ j .146 8 -2 362 . 2011. 0129 4. x .
Weaver, D.G. (2011) Internalization and market quality in a fragmented market structure. Social Science
Research Network. Available:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1846470.
Accessed: 18 September 2012.
Zhang, F. (2010) High-frequency trading, stock volatility, and price discovery. Social Science Research
Network. Available:
http://ssrn.com/abstract=1691679.
Accessed: 18 September 2012.
Zigrand, JP (2005) Rational asset pricing implications from realistic trading frictions. Journal of Business,
78(3): 871–892.
Zovko, I. and Farmer, D. (2002) The power of patience: a behavioral regularity in limit-order placement.
Quantitative Finance, 2(5): 387–392.
164
Annex C Glossary of terms and acronyms
Annex C Glossary of terms
and acronyms
Terms
The following denes the terms used in this Report.A list of acronyms is given at the end.
Algorithmic trading (AT) – different denitions of this term are used in various parts of the world. For
example, the following provides those used by the European Commission Directorate General Internal
Market and Services (DG MARKT) and the US Commodity Futures Trading Commission (CFTC). The
use of AT in this report is broadly consistent with both of these denitions:
DG MARKT: “Automated trading also known as algorithmic trading can be dened as the use of
computer programmes to enter trading orders where the computer algorithm decides on aspects
of execution of the order such as the timing, quantity and price of the order.”
1
CFTC: “The use of computer programs for entering trading orders with the computer algorithm
initiating orders or placing bids and offers.”
2
Bid-ask spread – the difference in price between the highest price that a buyer is willing to pay for
an asset and the lowest price for which a seller is willing to sell it. This difference in price is an implicit
trading cost to pay for immediate liquidity.
Broker crossing networks – trading platforms in which traders submit buy and sell orders and trades
are executed for matching orders at an exogenous price; typically the midpoint of the current bid
and offer spread. As there are often more orders on one side than the other, not all orders execute.
Trades that do execute, however, do so at a better price than they would have in the market.
Cloud computing – Ultra large scale data centres (vast warehouses full of interconnected computers)
which can be accessed remotely as a service via the internet, with the user of the remotely accessed
computers paying rental costs by the minute or by the hour.
Computer-based trading – this Project has taken a broad interpretation of computer-based trading
(CBT). A useful taxonomy of CBT was proposed in DR5, which identies four characteristics that can
be used to classify CBT systems. First, CBT systems can trade on an agency basis (i.e. attempting to get
the best possible execution of trades on behalf of clients) or a proprietary basis (i.e. trading using one’s
own capital); second, CBT systems may adopt liquidity-consuming (aggressive) or liquidity-supplying
(passive) trading styles; third, they may be classied as engaging in either uninformed or informed
trading; and, fourth, the trading strategy can be generated by the algorithm itself or, alternatively, the
algorithm is used only to optimally implement a decision taken otherwise.
Dark pool – trading systems in which buy and sell orders are submitted anonymously and are not
displayed to the public markets until execution. Broker crossing networks dened above are a type of
dark pools.
Designated liquidity provider – a general term given to a market participant who agrees to stand
ready to buy or sell an asset to accommodate market demand.
Exchange traded funds (ETF) – ETFs are securities traded on major exchanges as if they were standard
stock equities (shares in a company), but the ETF instead represents a share in a holding of assets, such
as commodities, currency or stock.
1 See European Commission (2010).
2 See CFTC (2010).
165
The Future of Computer Trading in Financial Markets
Financial market stability – the lack of extreme movements in asset prices over short time periods.
High Frequency Trading (HFT) – different denitions of this new term are used in various parts of the
world. For example, the following provides those used by European Securities and Markets Authority
(ESMA) and the Securities and Exchange Commission (SEC) in the US. The use of HFT in this report is
broadly consistent with both of these denitions:
ESMA: “Trading activities that employ sophisticated, algorithmic technologies to interpret signals
from the market and, in response, implement trading strategies that generally involve the high
frequency generation of orders and a low latency transmission of these orders to the market.
Related trading strategies mostly consist of either quasi market making or arbitraging within very
short time horizons. They usually involve the execution of trades on own account (rather than
for a client) and positions usually being closed out at the end of the day.”
3
SEC: “The term is relatively new and is not yet clearly dened. It typically is used to refer to
professional traders acting in a proprietary capacity that engage in strategies that generate a large
number of trades on a daily basis (…) Other characteristics often attributed to proprietary rms
engaged in HFT are: (1) the use of extraordinarily high speed and sophisticated computer programs
for generating, routing, and executing orders; (2) use of co-location services and individual data
feeds offered by exchanges and others to minimize network and other types of latencies; (3) very
short time-frames for establishing and liquidating positions; (4) the submission of numerous orders
that are cancelled shortly after submission; and (5) ending the trading day in as close to a at
position as possible (that is, not carrying signicant, unhedged positions over night).”
4
Informed or value-motivated traders – aim to prot by trading on the basis of information and
make use of information in news stories and related discussion and analysis to come to a view about
what price an instrument should be trading at either now or in the future, and then buy or sell that
instrument if their personal opinion on the price is different from the current market value.
Internalisation – the practice whereby some customer trades are executed internally by brokers or
intermediaries and so do not reach public markets.
Inventory traders – aim to prot by merely providing liquidity and act as ‘market-makers’: they hold
a sufciently large inventory that they are always able to service buy or sell requests, and they make
money by setting a higher price for selling than for buying. Inventory traders can, in principle, operate
protably without recourse to any information external to the market in which their instruments are
being traded.
Layering and spoong – layering refers to entering hidden orders on one side of the book (for
example, a sell) and simultaneously submitting visible orders on the other side of the book ( buys). The
visible buys orders are intended only to encourage others in the market to believe there is strong price
pressure on one side, thereby moving prices up. If this occurs, the hidden sell order executes, and the
trader then cancels the visible orders. Similarly, spoong involves using, and immediately cancelling, limit
orders in an attempt to lure traders to raise their own limits, again for the purpose of trading at an
articially inated price.
Liquidity – the ability to buy or sell an asset without greatly affecting its price. The more liquid the
market, the smaller the price impact of sales or purchases.
Locked or crossed market – a locked or crossed market is where the bid-ask spread is zero (negative)
and is evidence of poor linkages between markets and investors.
Long-only macro trading – long-only macro trading is where a trader maintains a portfolio of holdings
that are bought only in the expectation that their market value will increase (that is, the trader is taking
long positions, as opposed to short positions where the trader would benet if the market value
decreases); and where the trader’s alterations to the portfolio are driven by macroeconomic factors,
3 See ESMA (2011).
4 See SEC (2010).
166
Annex C Glossary of terms and acronyms
such as national interest rates, which tend to alter relatively slowly, rather than the second-by-second
uctuations commonly exploited by high frequency traders.
Market making – providing liquidity to buyers and sellers by acting as a counterparty. A market maker
buys from sellers and sells to buyers.
Order book – the collected limit orders to buy or sell an asset. Order books today are generally
electronic and allow traders to specify the prices at which they would like to buy or sell a specied
quantity of an asset.
Order ows – the arrival of buy orders and sell orders to the market.
Market abuse
5
– consists of market manipulation and insider dealing, which could arise from distributing
false information, or distorting prices and improper use of insider information.
Market abuse may be grouped into the following seven categories:
Insider dealing: when an insider deals, or tries to deal, on the basis of inside information
Improper disclosure: where an insider improperly discloses inside information to another person
Manipulating transactions: trading, or placing orders to trade, that gives a false or misleading
impression of the supply of, or demand for, one or more investments, thus raising the price of the
investment to an abnormal or articial level.
Misuse of information: is behaviour based on information that is not generally available but would
affect an investor’s decision about the terms on which to deal.
Manipulating devices: refers to trading, or placing orders to trade, employing ctitious devices or
any other form of deception or contrivance.
Dissemination: refers to giving out information that conveys a false or misleading impression about
an investment or the issuer of an investment where the person doing this knows the information to
be false or misleading.
Distortion and misleading behaviour: refers to behaviour that gives a false or misleading impression
of either the supply of, or demand for, an investment; or behaviour that otherwise distorts the
market in an investment.
Market efciency – the concept that market prices reect the true underlying value of the asset.
Market transparency – the ability to see market information. Post-trade transparency refers to the
ability to see trade prices and quantities. Pre-trade transparency refers to the ability to see quotes.
Momentum ignition – entry of orders or a series of orders intended to start or exacerbate a
trend, and to encourage other participants to accelerate or extend the trend in order to create an
opportunity to unwind/open a position at a favourable price.
Multilateral trading facility (MTF) – According to MiFID Directive, article 4, no. 15) a. MTF is “a
multilateral system, operated by an investment rm or a market operator, which brings together
multiple third-party buying and selling interests in nancial instruments - in the system and in
accordance with non-discretionary rules – in a way that results in a contract in accordance with
the provisions of Title II.”
6
Order anticipation strategies – a trader looks for the existence of large (for example) buyers, in the
objective of buying before these orders, in order to benet from their impact.
Order preferencing – arrangements whereby orders are sent to a prespecied exchange or broker/
dealer in exchange for a per share payment.
5 The source for denitions of market abuse related terms are EU Commission Staff Working Paper, (2011b) AFM, (Netherlands
Authority for Financial Markets), (2010).
6 See European Commission (2004).
167
The Future of Computer Trading in Financial Markets
Passive limit orders – these provide the counterparty for traders wishing to nd a buyer or seller in
the market.
Ping orders – entering small orders into a market in order to assess the real level of liquidity on that
venue (beyond that displayed).
Price discovery – the market process whereby new information is impounded into asset prices.
Price efciency – when an asset’s price reects the true underlying value of an asset.
Primary market – when a company issues equities (shares) to raise capital, that is the primary market
in action.
Quote stufng – entering large numbers of orders and/or cancellations/updates to orders so as
to create uncertainty for other participants, slowing down their process and to camouage the
manipulator’s own strategy.
Secondary market – when the shares from the primary market are subsequently traded among
investors and speculators, that is the secondary market in action.
Statistical arbitrage – also known as ‘stat arb’. One popular class of stat arb strategies identify long-
term statistical relationships between different nancial instruments and trade on the assumption that
any deviations from those long-term relationships are temporary aberrations and that the relationship
will revert to its mean in due course.
Suspicious transaction report (STR) – reports to competent authorities required under Article 6(9) of
the Market Abuse Directive where a person professionally arranging transactions reasonably suspects
that a transaction might constitute insider dealing or market manipulation.
The ‘touch’ – According to the Oxford English Dictionary, the touch is “the difference or spread
between the highest buying price and the lowest selling price in a commodities or share market.”
7
Trade-through – when a trade is executed at a price on one venue that is inferior to what was
available at another venue at the same time.
Transaction costs – the costs traders incur to buy or sell an asset.
Volatility – variability of an asset’s price over time, often measured in percentage terms.
168
7 See Oxford English Dictionary.
Annex C Glossary of terms and acronyms
Acronyms
AES automated execution system
AT algorithmic trading
ATS Alternative Trading System
BATS Better Alternative Trading System
CBT computer-based trading
CDO collateralised debt obligation
CDS credit default swaps
CFD contracts for difference
CFTC Commodity Futures Trading Commission
CLOB Central Limit Order Book
CME Chicago Mercantile Exchange
CoCo conditional convertible debt instrument
CPU central processing unit
CRA credit rating agency
DMA direct market access
EBS electronic broking services
EDHEC Ecole des Hautes Etudes Commerciales
EFDC European Financial Data Centre
EMEA Europe, the Middle East and Africa
EOFR European Ofce of Financial Research
ESMA European Securities and Markets Authority
ETF exchange traded fund
ETP exchange traded product
EU European Union
FESE Federation of European Securities Exchanges
FINRA US Financial Industry Regulatory Authority
FPC Financial Policy Committee
FPGA eld programmable gate array
FSA Financial Services Authority
FSOC Financial Stability Oversight Council
FX foreign exchange
GDP gross domestic product
GETCO Global Electronic Trading Company
HFT high frequency trading
HPC high performance computing
HRO high reliability organisation
IMF International Monetary Fund
IOSCO International Organization of Securities Commissions
IPO initial public offering
IPR intellectual property rights
ISE Istanbul Stock Exchange
ISO International Organisation for Standards
LEI legal entity identier
LIBOR London Interbank Offered Rate
LSE London Stock Exchange
LTI loan to income
LTV loan to value
MFR monthly ll ratio
MiFID Markets in Financial Instruments Directive
MQL minimum quote lifespan
MTF multilateral trading facility
NMS national market system
OBI order book imbalance
OECD Organisation for Economic Co-operation and Development
OER order-to-execution ratio
169
The Future of Computer Trading in Financial Markets
OFR
OTC
OTR
PC
PTP
RLP
RoE
SAM
SEAQ
SEC
SI
SMS
SOR
STP
STR
TAIFEX
TRF
TWAP
UNSCC
VWAP
Ofce of Financial Research
over-the-counter
order-to-trade ratio
personal computer
price/time priority
retail liquidity provider
return on equity
shared appreciation mortgages
Stock Exchange Automated Quotation
Securities and Exchange Commission
systematic internaliser
standard market size
smart order routing/router
straight-through processing
suspicious transaction report
Taiwan Futures Exchange
trade reporting facility
time-weighted average price
United Nations Standards Coordination Committee
volume-weighted average price
170
The Future of Computer Trading in Financial Markets Chapter Title
T U
Figure D.1: Project reports and papers
Final Report
Economic Impact Assessments Surveys and Interviews Workshop Reports
EIA1: Minimum resting times and order
message-to-trade ratio
SR1: Survey of end-users WRI: Chief Economists
workshop report
EIA2: Minimum resting times and transaction-
to-order ratios
IN1: Interviews with computer-
based traders
WR2: Industry workshop
report
EIA3: Minimum resting times, call markets
and circuit breakers
WR3: Drivers of change and
market structure
EIA4: Circuit breakers SC1: Consolidated scenarios
EIA6: Tick sizes
EIA7: Tick sizes
EIA8: Minimum obligations for market makers
EIA9: Circuit breakers
EIA10: Internalisation
EIA11: Continuous market vs. randomised
stop auctions and alternative priority rules
EIA12: Maker-taker pricing
EIA13: CLOB vs. exchange order books
EIA16: Algorithmic regulation Project reports and papers are
freely available to download at
http://www.bis.gov.uk/foresight
Note 1: Some report numbers
were initially allocated but were
not subsequently used.
Note 2: All driver reviews and
economic impact assessments
have passed a double blind peer
review process.
EIA17: Market abuse and surveillance
EIA18: Ratio of orders to transactions
EIA19: Minimum obligations for market makers
EIA20: Harmonised circuit breakers
EIA21: Economic impact of MiFID regulation
EIA22: Economic impact assessment of
proposal for CBT
The Future of Computer Trading in Financial Markets
FINAL PROJECT REPORT
Annex D Project
reports and papers
Driver Reviews
Executive Summary
The Future of
Computer Trading
in Financial Markets
An International Perspective
DR1: What has happened to UK equity market quality in
the last decade?
DR17: The evolution of algorithmic classes
DR2: Feedback effects and changes in the diversity of
trading strategies
DR18: Pricing liquidity in electronic markets
DR3: Technology trends in the nancial markets DR19: Market fragmentation in Europe: assessment and
prospects for market quality
DR4: The global nancial markets – an ultra-large-scale
systems perspective
DR20: Computer-based trading and market abuse
DR5: Computer-based trading, liquidity and trading costs DR21: High frequency trading and the execution cost
of institutional investors
DR6: An ecological perspective on the future of computer
trading
DR22: High frequency trading and end of day price
manipulation
DR7: Crashes and high frequency trading DR23: Algorithmic trading and changes in rms’ equity
capital
DR8: Automated analysis of news to compute market
sentiment
DR24: The impact of high frequency trading on market
integration – an empirical examination
DR9: Leverage, forced asset sales and market stability DR25: Exploring the ‘robot phase transition’ in
experimental human-algorithmic market
DR10: High frequency trading information and prots DR26: Computer trading and systemic risk – a nuclear
perspective
DR11: Impersonal efciency and the dangers of a fully
automated securities exchange
DR27: Brave new world: quantifying the new
instabilities and risks arising in sub second algorithmic
trading
DR12: High frequency trading and price efciency DR28: High frequency trading – assessing the impact on
market efciency and integrity
DR13: Studies of interactions between human traders and
algorithmic trading systems
DR29: Systemic risk arising from computer-based
trading and connections to the empirical literature on
systemic risk
DR14: Prospects for large scale nancial systems simulation DR30: Trust and reputation in nancial services
DR15: Impact of special relativity on securities regulation DR31: Standards in nancial services
DR16: Electronic trading and market structure
172
The Future of Computer Trading in Financial Markets Chapter Title
T U
Figure D.1: Project reports and papers
Executive Summary
The Future of
Computer Trading
in Financial Markets
An International Perspective
FINAL PROJECT REPORT
Driver Reviews
DR1: What has happened to UK equity market quality in
the last decade?
DR17: The evolution of algorithmic classes
DR2: Feedback effects and changes in the diversity of
trading strategies
DR18: Pricing liquidity in electronic markets
DR3: Technology trends in the nancial markets DR19: Market fragmentation in Europe: assessment and
prospects for market quality
DR4: The global nancial markets – an ultra-large-scale
systems perspective
DR20: Computer-based trading and market abuse
DR5: Computer-based trading, liquidity and trading costs DR21: High frequency trading and the execution cost
of institutional investors
DR6: An ecological perspective on the future of computer
trading
DR22: High frequency trading and end of day price
manipulation
DR7: Crashes and high frequency trading DR23: Algorithmic trading and changes in rms’ equity
capital
DR8: Automated analysis of news to compute market
sentiment
DR24: The impact of high frequency trading on market
integration – an empirical examination
DR9: Leverage, forced asset sales and market stability DR25: Exploring the ‘robot phase transition’ in
experimental human-algorithmic market
DR10: High frequency trading information and prots DR26: Computer trading and systemic risk – a nuclear
perspective
DR11: Impersonal efciency and the dangers of a fully
automated securities exchange
DR27: Brave new world: quantifying the new
instabilities and risks arising in sub second algorithmic
trading
DR12: High frequency trading and price efciency DR28: High frequency trading – assessing the impact on
market efciency and integrity
DR13: Studies of interactions between human traders and
algorithmic trading systems
DR29: Systemic risk arising from computer-based
trading and connections to the empirical literature on
systemic risk
DR14: Prospects for large scale nancial systems simulation DR30: Trust and reputation in nancial services
DR15: Impact of special relativity on securities regulation DR31: Standards in nancial services
DR16: Electronic trading and market structure
Annex D Project reports and papers
Final Report
Economic Impact Assessments Surveys and Interviews Workshop Reports
EIA1: Minimum resting times and order
message-to-trade ratio
SR1: Survey of end-users WRI: Chief Economists
workshop report
EIA2: Minimum resting times and transaction-
to-order ratios
IN1: Interviews with computer-
based traders
WR2: Industry workshop
report
EIA3: Minimum resting times, call markets
and circuit breakers
WR3: Drivers of change and
market structure
EIA4: Circuit breakers SC1: Consolidated scenarios
EIA6: Tick sizes
EIA7: Tick sizes
EIA8: Minimum obligations for market makers
EIA9: Circuit breakers
EIA10: Internalisation
EIA11: Continuous market vs. randomised
stop auctions and alternative priority rules
EIA12: Maker-taker pricing
EIA13: CLOB vs. exchange order books
EIA16: Algorithmic regulation Project reports and papers are
freely available to download at
http://www.bis.gov.uk/foresight
Note 1: Some report numbers
were initially allocated but were
not subsequently used.
Note 2: All driver reviews and
economic impact assessments
have passed a double blind peer
review process.
EIA17: Market abuse and surveillance
EIA18: Ratio of orders to transactions
EIA19: Minimum obligations for market makers
EIA20: Harmonised circuit breakers
EIA21: Economic impact of MiFID regulation
EIA22: Economic impact assessment of
proposal for CBT
173
The Future of Computer Trading in Financial Markets
Annex E Possible future scenarios
for computer-based trading in
financial markets
In the future, the development and uptake of computer-based trading (CBT) will be substantially
inuenced by a range of important drivers of change, acting alone and in concert. Regulation will be
particularly important
1
, as will technology development and roll-out
2
, but other key drivers identied in
this Project include
3
: demographic shifts, global economic cycles, geopolitics, changes in riskless assets,
changing asset classes, competition and changes in (dis)intermediation. Conversely, CBT will itself have
a profound effect on market operation, market structure and, in turn, feed back to inuence some of
those key drivers of change.
While the time horizon of this Project is just ten years or so, there are considerable uncertainties
inherent in many of these drivers of change. For this reason, it was impossible to determine a single
‘most probable’ future scenario. Therefore, this Project has developed four plausible future scenarios –
in essence these sample the future ‘possibility space’
4
.
While the development of these scenarios has not substantially affected the detailed assessments
of individual regulatory measures in Chapter 6, the scenarios generated were considered useful in
stimulating thinking about the wider context in which CBT operates. For this reason, the following
provides a brief overview of the four scenarios considered.
Four future scenarios
There are many ways in which scenarios for future nancial markets can be cast. However, as
computer trading was a central concern, two broad drivers of change were considered particularly
important following the workshops mentioned above. One driver is the extent to which future
growth is sustained and convergent and the other is the extent to which markets are open and reliant
on decentralised technologies. These are therefore represented as the horizontal and vertical axes
respectively in Figure E.1. The quadrants of this gure therefore dene four contrasting scenarios,
described briey below.
1 See Chapter 6.
2 See Chapter 2.
3 These drivers of change were identied in workshops in London, Singapore and New York involving leading industry
practitioners, and a workshop of chief economists in London. See Chapter 1 for a brief description of these drivers and the
various workshop reports. See WR1, WR2, WR3, SC1 (Annex D refers).
4 Each of the workshops mentioned emphasised different drivers of change as particularly important (see Figure 1.1 in Chapter 1)
although regulation and demographic change emerged as consistently important. The scenarios presented here are not
therefore representative of any one workshop, but rather an amalgamation of views expressed.
174
Annex E Possible future scenarios for computer-based trading in nancial markets
Figure E.1: The four possible future scenarios, illustrated as quadrants spanning a
two-dimensional space where one dimension is the extent which future growth
is sustained and convergent (horizontal axis) and the other is the extent which
markets are open and reliant on decentralized technologies (vertical axis)
Scenario D – Geopolitical and
economic competition
• Competing world powers and
economic models
• Light touch regulation but low growth
• Investment shifts to emerging markets
• Companies in much of Asia owned by
state and nancial elite
• In Asia, retail trading of derivatives and
other synthetics explodes – copied in
the West
• Bank shrinkage and deleveraging
• Real assets preferred to nancial
• Shadow banking thrives; households
SMEs suffer
• New instruments and online
exchanges for company nancing
Sluggish and
fragmented
growth
Scenario C – Retrenchment and
global tensions
• Global lost decade – Economic
systems worldwide retrench in the
face of severe challenges
• Tensions rise across the globe – war?
• Bank failures, restructurings,
nationalisation
• Capital controls, protectionism
• Dollar loses reserve currency status
• Pressure on exchanges, trading rms,
leads to consolidation, rebundling,
monopolies
• Proliferation of synthetic products
with opaque structures, responding to
demand for ‘safe returns’
• HFT grows. Churning and copycat
strategies “are adopted”
Open
markets,
distributed
technologies
Closed
markets,
centralised
technologies
Scenario A – Back to the 1980s
• Globalisation advances
• Global imbalances resolved: US decit
stabilise; China growth stabilises
• Increased nancial market activity
(nance increases share of GDP)
• Contained EU defaults
• Competition leads to technology
innovation, pressure on traditional
trading margins
• Tech-literate new trading generation
• New platforms, products
• Interlinked global exchanges, single
exchange view
• Competing trading and clearing
components marketed to exchanges
Sustained and
convergent
growth
Scenario B – Back to the 1950s
(post-war boom)
• More regulation, growth remains high
• Emerging economies take the lead
• International institutions play an
increasing role in governing economic
and political systems
• Rebalancing of capital and trading
volumes to emerging markets
• Financial markets recongure along
regional lines
• Regional exchanges dominate,
interconnected in a carefully regulated
system
• Less complex assets
• Responding to low beta margins,
more macro strategies lead to
correlation, lower volumes
175
The Future of Computer Trading in Financial Markets
Scenario A – Back to the 1980s
In this scenario, global imbalances begin to be resolved as the savings and investments imbalance
between East and West reduces. Globalisation makes steady progress; growth returns to the
developed world and picks up again in the emerging economies.
The Japanese economy develops (modest) surpluses, US decits stabilise, Europe survives a series of
‘corrective’ defaults without major disruption and China’s growth is stabilised by the emergence of
healthy levels of domestic demand.
Financial markets account for an increasing proportion of GDP, although equities lose market share
to derivatives and other synthetic products. Increasing amounts of investment and trading activity are
accounted for by hedge funds, in part to avoid barriers and regulations. New asset classes emerge, such
as water futures and country insurance bonds.
Novel platforms and nancial services are also developed: for example, software providers sell
competing trading and clearing components to exchanges and also offer retail versions of their
products, such as algorithmic trading on mobile handsets. Retail investment volumes increase strongly.
The interconnection of global exchanges continues, offering a single view of multiple exchanges and
opening the way to a ‘nancial grid’ system. The market for trading intermediaries attracts new
entrants; competition leads to technology innovation and pressure on traditional trading margins.
Overall in this scenario liquidity is good, nancial stability increases, volatility declines, prices efciently
incorporate all knowledge and risks, transaction costs are low and markets effectively self-regulate
against opportunistic or predatory behaviour.
Scenario B – Back to the 1950s (post-war boom)
In this scenario, markets have become less open and nance is increasingly regulated around the world.
Capital and trading volumes are ‘rebalanced’ in favour of emerging markets, which are better able to
sustain growth, while Western economies remain encumbered by their debt burdens. International
institutions play an increasing role in governing economic and political systems, offering a greater voice
to the new economies. The Washington consensus of lightly regulated nancial markets and capital
ows has passed.
Financial markets are recongured along regional lines and investment is biased towards domestic
markets. Governments use regulation to capitalise on – and protect – local advantages. Globalisation
declines, but does not disappear altogether; trade has reached a plateau; some regional trading blocs
(such as the EU) remain effective. Trade deals tend to be bilateral rather than multilateral, and more in
the interest of the new economies.
The UK seeks to rediscover the macroeconomic policy stance of the post-war years, when higher levels
of nancial market regulation coexisted with annual growth rates of 5% or more. Such growth rates
prove out of reach in the 2010s however, as debt is paid down.
Protectionist government intervention means nancial resources are not allocated efciently; there is
more ‘strategic’ allocation and less short-term prot maximisation. Price discovery is distorted; risk is
not efciently priced. Savings in the emerging countries are reinvested there; the West has to bid over
the odds to attract investment capital. Political support for free trade and open markets weakens in the
West, leading to a gradual departure of entrepreneurial talent.
Institutional investment outweighs retail investment; nominal interest rates do not equalise on a global
basis. Large regional exchanges dominate; smaller rms and platforms are perceived as risky after
well-publicised failures. Stock exchange regulation privileges consumer protection. There are more
manufacturing jobs and the nancial services industry accounts for a smaller proportion of GDP.
176
Annex E Possible future scenarios for computer-based trading in nancial markets
The nancial world sees a return to less exotic products and services, epitomised by the return of the
building-society model. There is more interest in risk sharing than aggressive (nancial) gain seeking.
Continued high frequency trading (HFT) use leads to reduced trading margins for institutions, while
macro strategies lead to correlation thus reducing activity levels.
As this scenario entails compartmentalisation, asset simplication and nancial restrictions, especially
of international ows, liquidity and volatility will generally reduce, but nancial stability will increase.
Markets will not price properly for risk, which will harm price efciency and discovery and increase
the volatility of real returns to specic assets. Unit transaction costs will increase. Market integrity will
decline but market insiders and those who circumvent the markets will do well.
Scenario C – Retrenchment and global tensions
In this scenario, the collapse of the Euro and ensuing depression leads to the reintroduction of capital
controls in Europe; banks are nationalised amid a resurgence of protectionism. Economic systems
worldwide retrench in the face of severe challenges.
The strengthening US dollar loses its reserve currency role as the USA attempts to improve its current
account decits by devaluation (rather than reform of scal policy and savings–investment imbalances),
triggering ination and increasing the perceived riskiness of both US sovereign debt and other dollar-
denominated assets. The resulting confusion and the lack of an alternative ‘risk-free’ reference asset,
lead to violent and unpredictable price uctuations.
Globally, there are persistent increases in decits and a Japanese-style lost decade lies ahead. The
West reacts by blaming others and turning protectionist. There are bank failures, restructurings and
nationalisations.
The loss of equilibrating global trades and the resulting restrictions on free movement of assets and
entry exacerbate inequality within and among nations. This is extended by the failure of markets
properly to price claims in different dates and states.
Populist politicians and parties come to the fore on a wave of discontent. ‘Arab spring’ turns into
‘Arab winter’; developed and emerging nations blame each other for their growing economic troubles;
China and India ght for access to the remaining world markets. These tensions are set to escalate
dangerously.
Exchanges are hit by falling volumes as people lose condence in markets and seek liquidity. Several
major trading rms fail. This leads to concentration; surviving rms use their dominance to control
access to exchanges and ‘re-bundle’ elements of the trading and clearing process.
Responding to demand from pension funds and the withdrawal of support from central banks, liquidity
provision is increasingly privatised through ‘liquidity transfers’, spreading liquidity risk around the
banking system.
Structured products are used to offer ‘safe’ returns to the retail market. ‘Alpha’ trading strategies –
which aim to achieve returns that beat a relevant benchmark – require taking on increasing principal
risk. HFT grows, as returns are desperately sought. Churning and copycat strategies are adopted, and
multiple feedback loops increase the likelihood of a catastrophic nancial event.
As in Scenario B, this scenario involves widespread use of capital controls and a strong ‘home bias’. In
contrast, however, it has limited or no growth. In consequence, although liquidity is reduced, increases
in nancial stability cannot be taken for granted, because low growth threatens the existence of a
riskless asset. On the other hand, short of catastrophe, low growth depresses volatility and assets
prices alike. Transaction costs are up. The pervasiveness of inefcient barriers to economic activities
produces corruption and other threats to market integrity.
177
The Future of Computer Trading in Financial Markets
Scenario D – Geopolitical and economic competition
In this scenario, regulation on the whole remains ‘light touch’, but growth still falters. As stagnation
persists, however, some nations opt for greater regulation and restrict access to their nancial markets.
World powers have entered a phase of geopolitical and economic competition.
Western economies stagnate; quantitative easing has become less effective. Even China sees GDP
growth fall, but the ‘Asian model’ of a state-dominated economy remains in the ascendant. Companies
in much of Asia are owned by the state and nancial elite – share ownership does not converge with
patterns in the West. The euro, sterling and yen lose value, and the US dollar shows signs of weakness.
More trade is carried out in renminbi.
The US deleverages, in part by renegotiating or defaulting on debts. Underemployment has lowered
productivity growth, while differing skill levels exacerbate wage inequality. Rising inequality has
increased political fragmentation and extremism.
In countries that persist with light-touch regulation and open markets, banks and some other nancial
institutions have shrunk. This is partly a consequence of competition, but also to do with deleveraging.
The ight of capital to emerging – and liberalising – markets prolongs the stagnation of the developed
economies.
Investors have become desperate for yield. ‘Buy and hold’ strategies and pension funds underperform.
As a result, capital ows from the USA to China, Latin America and other emerging markets, and
from traditional nancial assets to ‘real’ investments; some markets struggle to handle these shifts and
episodes of illiquidity occur.
The shadow-banking sector does well, but households and small rms struggle for access to capital.
New instruments and online exchanges are created for company nancing, as traditional capital-raising
falters.
Seeking new sources of revenue in a stagnant trading environment, rms in the West offer trading apps
to their customers, having seen them take off among Asian retail investors. Retail trading of derivatives
and other synthetics grows strongly, though many see this as a betting culture taking hold in the
absence of reliable yields from traditional investments.
Problems considered less urgent, such as the environment, take a back seat to short-term economic
policy. Domestic and international political tensions are stoked by migration, continued economic
malaise and growing differences in economic performance.
Overall, this scenario embodies openness without strong global growth. This exacerbates unstable
global nancial ows. Liquidity is unreliable, nancial stability is threatened by a series of speculative
bubbles and crashes and volatility increases. The unstable ight of capital through a fragmented nancial
system indicates effective price discovery and low transactions costs. The zero-sum nature of this
scenario makes the threats to market integrity even worse than in Scenario C.
178
Printed in the UK on recycled paper with a minimum HMSO score of 75
First published October 2012.
The Government Office for Science
© Crown copyright. URN 12/1086
