Market Fairness: The Poor Country Cousin of Market Efficiency

J Bus Ethics (2018) 147:5–23
https://doi.org/10.1007/s10551-015-2964-y
Market Fairness: The Poor Country Cousin of Market Efficiency

Michael J. Aitken1 • Angelo Aspris2 • Sean Foley2 • Frederick H. de B. Harris3
Received: 14 November 2014 / Accepted: 14 November 2015 / Published online: 8 December 2015
Springer Science+Business Media Dordrecht 2015
Abstract Both fairness and efficiency are important Keywords Market quality Market fairness
considerations in market design and regulation, yet many Manipulation Information leakage Algorithmic trading
regulators have neither defined nor measured these con-
cepts. We develop an evidencebased policy framework in JEL Classification G28 Financial Institutions and
which these are both defined and measured using a series of Services Government Policy and Regulation
empirical proxies. We then build a systems estimation
model to examine the 2003–2011 explosive growth in
algorithmic trading (AT) on the London Stock Exchange Introduction
and NYSE Euronext Paris. Our results show that greater
AT is associated with increased transactional efficiency Implementation of the Markets in Financial Instruments
and reduced information leakage in top quintile stocks. For Directive (MiFID1) in 2007 markedly altered the opera-
less liquid stocks, manipulation at the close declines. We tions of European exchanges greatly facilitating competi-
also document the tradeoff between reduced spreads and tion from the multilateral trading facilities (MTFs) such as
increased manipulation or information leakage following Chi-X. Like the concurrent Reg NMS in America, MiFID1
the introduction of MiFID1. accompanied a substantial increase in the level of algo-
rithmic trading (AT) that was evolving on its own. We
investigate how the growth of AT impacted the market
quality of two exchanges—the London Stock Exchange
(LSE) and NYSE-Euronext-Paris (Euronext). Our contri-
Electronic supplementary material The online version of this
article (doi:10.1007/s10551-015-2964-y) contains supplementary bution hinges upon an expansive definition of market
material, which is available to authorised users. quality that includes not only the usual bid-ask spread and
price volatility, traditional proxies for transaction cost and
& Frederick H. de B. Harris
[email protected]
informational efficiency, but also our recently demon-
strated new proxies for market fairness based on alerts
Michael J. Aitken
[email protected]
generated by market surveillance systems (Aitken et al.
2014).
Angelo Aspris
[email protected]
According to the IOSCO 2003 statement on the Objec-
tives and Principles of Securities Regulation,1 security
Sean Foley
[email protected]
market design changes should pass the tests of both
1 1
Macquarie University, Sydney, Australia See Foreword and Executive Summary of Objectives and Principles
2 of Securities Regulation, May 2003, IOSCO which identifies the key
School of Business, Discipline of Finance, University of
objective of securities regulation as ‘‘ensuring that markets are fair,
Sydney, Sydney, Australia
efficient and transparent (emphasis added)’’. For discussions of why
3
School of Business, Wake Forest University, Winston-Salem, the third concept transparency is a means to an end rather than an end
USA in itself, see Aitken and Leduc (2012).
123
6 M. J. Aitken et al.
fairness and efficiency. In line with this dual objective,

Aitken and Harris (2011) propose a market quality
framework in which the five key elements of market design
(i.e. technology, information, participants, regulation and
instruments) are related to metrics representing market
fairness (i.e. less market manipulation, insider trading and
broker-agency conflict) as well as market efficiency (i.e.
transaction costs and price discovery)—see Exhibit 1.2
Because of the data and reporting infrastructure required,
market fairness has been largely ignored relative to market
efficiency. The present study demonstrates how to imple-
ment the Aitken–Harris market quality framework by
analysing dislocation of the end-of-day price (a proxy for
market manipulation), information leakage prior to price-
sensitive announcements (a proxy for insider trading) and
bid-ask spreads (a proxy for transaction cost efficiency).
We then provide a systems estimation model for assessing
the simultaneous effect of AT on these proxies for fairness
and efficiency.
In the case of trading ahead of price-sensitive
announcements, hedge funds and prop desks today con-
stantly monitor the state of the market stock by stock and
decompose quickly the permanent from the transitory
shocks to the order flow.3 As a result, the footprints of an
insider are now more quickly detected. AT then allows
the full price impact of new information to occur very
quickly (Hendershott and Riordan 2013), thereby reducing
the profitability of those who would seek to trade ahead of
Exhibit 1 a Market quality framework and b security market quality
the public release of such information. Hence, we
hypothesize and investigate whether increased AT reduces
information leakage, thereby increasing the fairness of the investigate whether increased AT reduces end-of-day
marketplace. manipulation, thereby increasing the fairness of the mar-
Market manipulators attempt to dislocate the full-in- ketplace. Finally, AT may also reduce the transactions
formation efficient price by deceiving partially informed costs of executing a trade, thereby increasing market
investors into believing that a new permanent price trend is efficiency.
underway and then profit from unwinding positions before Decomposing market quality into fairness and efficiency
the ramped price reverts.4 However, very low latency AT and allowing for interaction between them, potentially
(executing within milliseconds) often allows hedge funds offers great insight into the impact of market design
and proprietary trading desks (where brokerage firm rather changes on the published mandate of regulators to address
than client interests and funds are at risk) to unwind the dual objectives of market fairness and efficiency.
positions faster than the manipulators themselves. As a
result, ramping violations at the close are now much less The Market Quality Framework: An Overview
profitable than before AT. Hence, we hypothesize and
Stakeholders within the financial community (regulators,
2 exchanges and financial intermediaries) motivate changes
The authors define an efficient market as ‘‘a market in which it is
cheap to trade and the prices at which one is trading reflect all in market design. Recent high profile changes include
available information’’. We define a fair market as ‘‘…a market in algorithmic and high-frequency trading, co-location of
which prohibited trading behaviours are minimised’’. proprietary trading engines and exchange servers, bans on
3
For more explanation of the role of permanent and transitory shocks short sales, anonymous client services in dark pools,
in the execution of trading strategies, see Chakravarty et al. (2009) transaction taxes, trading infrastructure cost recovery fees,
and the Appendix to Harris et al. (2014).
4 order speed delays, as well as RegNMS and MiFID1-fa-
Dugast and Foucault (2014) develop a model of price reversals as
speculators trade fast on false signals and then revert back after cilitated dispersion of the order flow. Regulators are
processing the signal. required to sign off on these changes by ensuring they
123
Market Fairness: The Poor Country Cousin of Market Efficiency 7
enhance both fairness and efficiency or at a minimum harm implications from using this market quality (MQ) frame-
neither.5 This is particularly important given that the work. On the fairness/integrity side of the MQ ledger in
simultaneous equations estimation model introduced in this Exhibit 1, Panel B, we address first the dislocation of the
paper allows for the possibility of trade-offs between closing price with trade-based manipulation at the close
fairness and efficiency, the first evidence of which was (MTC) achieved through ramping or churning or squeezes.
outlined in Aitken et al. (2014). Although much prior These trading behaviours are distinguished from market
research has been done on the impact of market design manipulation due to false rumours or misleading partial
change on market efficiency, little or no work has been disclosures. In addition, we have developed a metric for
done on market fairness. Market fairness is the poor information leakage related to insider trading which relies
country cousin of market efficiency. So why is that? First, upon statistically exceptional returns from trading ahead of
the data requirements are daunting; just the publicly price-sensitive announcements. Without the release of cli-
available data on individual transactions are massive. The ent ID data (which brokers routinely provide to most
Security Industry Research Centre of Asia–Pacific modern exchanges), estimating the incidence of broker-
(SIRCA) expended more than $50 million to build the client conflicts such as front running is impossible.
infrastructure necessary to distribute consistent tick orders On the efficiency side, the implicit transaction costs of
and counterparty trade confirmations for 100? exchanges. executing a trade are measured in the traditional way with
Second, the audit trail data, required to prove insider bid-ask spreads and price impact.8 Finally, the MQ
trading and some types of market manipulation, are gen- framework operationalises price discovery concepts of
erally not available to researchers.6 After first building the informational efficiency with three time-series metrics—
necessary data infrastructure (in conjunction with SIRCA), permanent information impounding (PII) and price dis-
the MQ modelling framework has emerged over the last covery efficiency (PDE) of an execution channel (Harris
couple of years at the Capital Markets Cooperative et al. 2002, 2014), as well as orthogonal impulse response
Research Centre (CMCRC).7 Exhibit 1, Panel A displays functions to measure market resiliency (Chakravarty et al.
on the far left the five pivotal elements of market design 2009).
(Technology, Regulation, Information, Market Participants The proposed fairness-efficiency framework for assess-
and Instruments) that together impact two dimensions of ing market quality differs in several important respects
market efficiency (Transaction Costs and Informational from previous work on prohibited trading behaviours. Kyle
Efficiency) as well as three dimensions of market fairness and Vishwanathan (2008) define illegal price manipulation
(Market Manipulation, Insider Trading and Broker-Agent with reference to proxies for security market efficiency
Conflict) all of which we contend should be measured to alone—namely, price discovery and liquidity. Hasbrouck
provide a full assessment of the impact of any market and Saar (2013) conclude that increased AT improves
design change on market quality. short-term price volatility, spreads and displayed depth.
Both Brogaard (2010) and Jarnecic and Snape (2010) find
Fairness-Efficiency Objectives of Security Market Design that higher HFT results in a reduction in the bid-ask spread.
and Regulation Hendershott et al. (2011) find that higher AT lowers both
the effective and the realised spread. In contrast, Hender-
Aitken and Harris (2011) elaborate upon the rationale for, shott and Moulton (2011) show that lower latency trading
implementation of, exemplary proxies and the policy systems on the NYSE increase spreads but reduce noise,
making prices more efficient.
5
In sum, no one is quite sure as yet whether the onslaught
See for example the statements of objectives on websites of the
of HFT/AT is good or bad for market quality. AT and HFT
ASIC (Australia): a fair and efficient market characterised by
integrity and transparency; the SEC (US): fair, orderly and efficient have reduced the size available at the best bid and offer.
markets; the FSA (UK): efficient, orderly and fair markets; and the New internal crossing networks of undisplayed liquidity
OSC (Canada): fair and efficient capital markets [emphases added]. purportedly favour HFT at the expense of institutional
6
The same goes for broker-client conflicts of interest where brokers traders who would like to execute large block trades. To
are seldom required to add a condition code identifying whether the
avoid front running, institutions must now employ
order/trade is as principal or as agent. This information became
available in Australia from July 28, 2014, but there is a $1 m fine for increasingly sophisticated slice-and-dice mechanisms to
providing access to the data outside a regulatory or exchange inquiry. execute size but with an attendant substantial increase in
7
The founder of SIRCA and CEO of CMCRC is the first co-author
on this paper. CMCRC is a research consortium of universities, bank/
8
brokerage houses, national regulators and exchange operators world- We acknowledge that fees for clearing and settlement (i.e. the
wide. For more description on the Market Quality Dashboard go to explicit transaction costs) have become increasingly important in
https://www.youtube.com/watch?v=NegVKCHmsGw&list= some of our research because their magnitude now exceeds that of the
UUnyU9-4WAduJhlYOGTHTf3w. bid-ask spread in all the largest markets worldwide.
123
explicit transaction costs. Post-Reg NMS and MiFID1, the surveillance procedures and investigatory rules in numer-
incentive of market makers has fallen sharply to monitor ous and often unobserved ways thereby introducing con-
this now widely dispersed order flow enough to frequently current shocks that affect the incidence of manipulation
update their quotes, potentially harming informational and insider trading.
efficiency. In order to enable outcomes-based policy mak- Our simultaneous equations research reveals a fairness
ing, we propose a broad-based assessment of not only the and efficiency trade-off inherent in the six-fold increase in
efficiency but also the fairness implications of AT. AT over the period 2004–2011 in both London and Paris.
Although effective spreads declined in top decile stocks in
International Comparisons of Market Fairness both markets,9 in less liquid stocks, spreads have increased
and Efficiency in Paris. Crucially, our results also indicate, however, a
declining incidence of manipulation in both markets in all
Across 21 leading equity markets worldwide, the price deciles as AT increases. Information leakage alerts asso-
impact metric of the resiliency of a market in absorbing ciated with insider trading decline as well in both markets
buy and sell shocks is 10-fold greater from the most to the but only in the most liquid stocks. Policy decisions about
least efficient worldwide (Aitken and Harris 2011). Our AT such as signal delay mandates (SEC), cost recovery
2010 estimates of price impact (the percentage price fees (LSE) and transaction taxes (Euronext-Paris) require
change from larger trades) vary from 7 to 10 basis points unbiased highly precise estimates of this fairness-efficiency
(0.07–0.10 of 1 %) for NASDAQ and NYSE-Euronext up trade-off.
to 21 b.p. for the Australian Stock Exchange (ranked 10th), By introducing and calibrating the effect of market
and 69 basis points for the least efficient market. Across all design change on two elements of market fairness as well
markets we have studied, the average price impact is 24 as on effective bid-ask spreads, we make four separate
basis points. contributions to the existing literature on AT.
Moreover, metrics of market fairness exhibit an even
(1) We are the first to trace the effects of AT on market
wider range. The incidence of trading ahead of price-
manipulation and insider trading.
sensitive announcements, for example, varies from 0.02 to
(2) We are the first to address reverse causality from
0.44 of 1 % of total turnover, and cumulative abnormal
fairness violations onto AT.
profit from such insider trading varies from as little as
(3) By estimating a system of 3SLS simultaneous
$2556 per security per year up to $21 million (Aitken and
equations, we are able to incorporate the cross-
Harris 2011). Across the top ten (lowest information
equation correlation of residual disturbances in the
leakage) markets, the 2003–2009 average incidence of
order flow. We find that these residuals which we
trading ahead of price-sensitive announcements ranges
interpret as technology shocks and information
from 0.02 of 1 % for NYSE-Euronext, to 0.067 of 1 %
arrivals about the ‘‘state of the market’’ do in fact
for Hong Kong. The wide range of price impact metrics
influence concurrent manipulation, information leak-
and trading ahead of price-sensitive announcements
age and spread decisions in complex non-obvious
reflects in part constant innovation as markets compete to
ways.
attract order flow with new technologies, instruments,
(4) We are able to isolate and control for changes in
participants, information structures and regulatory
market design across exchanges and over time that
environments.
would otherwise confound the estimated effects of
AT. In this paper, we highlight two such market
design differences over the 2004–2011 sample
Contribution of Our Market Fairness Research
period: first, the imposition of MiFID1 regulations,
and second, the contrast between LSE’s hybrid
In this paper, our major contribution is to demonstrate for
market with its worked principal agreements
the first time the simultaneous effects of fairness and effi-
(WPAs) by London broker-dealers intertwined with
ciency on security market quality. With manipulation at the
an electronic limit order book versus the NYSE-
close (MTC) and trading ahead of price-sensitive
Euronext’s fully electronic limit order book supple-
announcements (LEAKAGE) affecting liquidity and there-
mented by liquidity providers.
fore SPREADS, system estimation techniques are required
to remove simultaneity bias with precision. Consider a
change in latency technology affecting the incidence and
cancellation strategies of algorithmic traders. These AT 9
In June 2009, one of the principal MTFs facilitating AT, Chi-X,
developments are likely to trigger the exchange operators, accelerated this gain in liquidity and transactional efficiency by
who are self-regulated organizations (SROs), to alter their sharply reducing settlement fees.
123
Institutional Context, Definition of AT, marketplace implanted by MiFID1, did achieve lower
and Related Theory Literature transaction fees as well as lower spreads.
Markets in Financial Instruments Directive 2007 What Exactly is Algorithmic Trading?

(MiFID1)
AT trading decisions are pre-designed by trade optimisa-
The implementation of MiFID1 in 2007 transformed the tion search engines, and submissions are automated and
European financial markets. Though not designed to pro- executed without human intervention. AT is employed by
mote AT/HFT, the new rules regarding best execution, proprietary traders to accomplish rapid order submissions
market transparency and market organisation resulted in and cancellations, often accentuated by co-locating HFT
this outcome. At the exchange level, infrastructure devel- servers beside an exchange. The dramatic growth in AT is
opment to reduce latency supported increased AT/HFT. attributable to lower latency (order routing speed) and the
For example, in 2011 the LSE introduced MillenniumIT, substantially reduced explicit trading costs on multilateral
which reduced latency to less than 120 ls, making LSE trade facilities such as Chi-X. A broad range of AT trading
one of the fastest trading venues in the world. Later that strategies is employed from pseudo market-making to
same year, the LSE Group fully acquired Turquoise whose pinging for pockets of undisplayed liquidity to statistical
platform achieves latencies of 90 ls. The already sub- arbitrage by quantitative hedge funds (Kumar et al. 2011).
stantial volume of AT/HFT in London soon exploded.10 Among the participants in AT are high-frequency traders
MiFID1 introduced four other key changes with impli- (HFT) who on average lose money on positions taken for
cations for AT/HFT: more than 5 s (Menkveld 2013). Instead, HFT realise
profits in a few milliseconds from small multi-venue price
1. The abolition of the concentration rule,
deviations in hundreds and sometimes thousands of intra-
2. The establishment of best execution obligations qual-
day transactions.
ified by model building,
AT is distinguished from pure low latency trading by
3. The encouragement of greater pre- and post-trade
HFT which seeks to profit on a fleeting informational
transparency,
advantage in acquiring and sending faster message traffic
4. And the introduction of passporting for security
to exchange servers without optimisation or real-time
clearing and settlement.
forecasting involved. Thus, HFT is a subset of a broader
MiFID1’s abolition of the concentration rule requiring class of AT. AT also employs fast order cancellation and
trades to go through a regulated exchange ensured greater replacement to arbitrage between competing execution
competition among trading venues. Best execution obli- channels but does so based specifically on real-time fore-
gations allowed new entrants such as the MTF Chi-X and casting models of the stock-specific ‘‘state of the market’’.
the systematic internaliser Merrill Lynch to attract order Some proprietary trading desks earn as much as $1200 per
flow from the incumbent exchanges who were thereby day per stock offering this forecasting service to large
forced to compete on explicit as well as implicit transaction insurance company and mutual fund clients. In addition,
costs. Settlement and clearing fees became a key feature of AT is routinely employed in agency trading of large
pricing policy first on the MTFs and later on the exchanges institutional orders where it is designed to stealthily capture
themselves (Harris et al. 2014). Market operators began to liquidity, to reduce order instruction footprints or simply to
offer alternative fee structures that were typically asym- minimize implementation shortfall from failing to com-
metric. For example, LSE offered a lower clearing and pletely execute an order instruction for any given price
settlement fee (and in some circumstances a rebate) to impact.
market participants who provided liquidity relative to those In sum, although many exchanges identify AT and HFT
participants who demanded liquidity. As expected, Gomber slightly differently, HFT generally refers to servers with
et al. (2011) and Harris and DiMarco (2012) demonstrate message traffic several orders of magnitude more frequent
that the more dispersed, yet evidently more harmonised than trade executions, that tend to execute positions held
only intraday, and locate so as to minimize latency. The
surveillance alerts of self-regulated organizations (SROs)
10
The estimated incidence of HFT varies widely across countries. and follow-on regulatory investigations are triggered by
Brogaard’s (2010) analysis of U.S equities finds that 60 % of all abnormal price movements sometimes traceable to such
NASDAQ trades involve an HFT on at least one side. Using a 2009 HFT servers. Again, HFT is a subset of AT; not all com-
sample for LSE, Jarnecic and Snape (2010) estimated that 40 % of
puter-determined trading operates through HFT servers.
trades include an HFT. In their 2010 response to the Committee of
European Securities Regulators (CESR), the LSE estimated that HFT For example, institutional trading desk instructions often
represented 33 % of total UK equities trading initiate computerised execution programs that scan
123
multiple platforms to ‘slice and dice’ large portfolio remaining level 10 quote updates into one of the following
rebalancing orders, to execute pairs trades or to conduct categories:
hedging strategies. These trades (and the price patterns they
1. New limit order that increases depth (at any of the 10
generate) are quickly shown to be artefacts of complex AT
levels),
strategies not demonstrative of illegal intent.
2. Amendment: adds to depth at some levels and reduces
depth at others,
AT/HFT Theory and Empirical Testing
3. Cancellation orders that only reduce depth.
In a theoretical model of HFT executed through an elec- In defining our proxy measure of AT, we focus on
tronic limit order book, Cvitanic and Kirilenko (2010) cancellation orders rather than total message traffic.
show a lowered duration between trades, reduced average Specifically, the cancellations from item 3 are compared to
trade value and reduced price volatility. A key assumption the number of trades to form a cancel-to-trade ratio
was that high-frequency traders act as uninformed market (CTR).12 As it turns out, upon careful examination, the
makers. In contrast, Jovanovic and Menkveld (2011) order-to-trade ratio (OTR) is exogenous in Hausman
assume ATs are more informed than their counterparts. specification testing whereas CTR is not. Therefore, we
Informed ATs would reduce spreads since they are better endogenise CTR as the left-hand-side metric of AT which
able to detect stale orders and avoid adverse selection. we identify with OTR and other instruments pretested for
Therefore, on the one hand, AT improves informational both their strength and exogeneity (see the online Techni-
efficiency. However, the enhanced speed of execution cal Appendix—Specification and Pooling Pretests).
reduces the willingness of human participants to enter the To determine the number of order cancellations, we
market, reducing liquidity, widening effective spreads and match trade executions against the limit order book
thereby harming transaction cost efficiency. Hendershott immediately prior to trades. When an incoming mar-
et al. (2011) and McInish and Upson (2013) test these ketable order executes against more than one standing limit
propositions and find that AT increases the realised spread order, multiple messages are generated for each standing
earned by liquidity providers relative to slow traders not limit order. Following Hasbrouck and Saar (2013), we
employing AT.11 combine into a single order all marketable order arrivals
Zhang (2010) argues that in theory greater HFT worsens within the same millisecond and in the same direction that
price discovery by focusing price adjustment dynamics too are unbroken by any non-execution message.
often on transitory shocks reflecting order imbalance as Exhibit 2 shows the six-fold increase in CTR from the
opposed to permanent price moves reflecting changes in onset of MiFID1 in November 2007 to mid-2010. By
fundamental value. But using Gonzalo-Granger’s time- facilitating the competitive dispersal of the order flow and
series decomposition techniques (Yan and Zivot 2010), introducing a passport rule, MIFID1 was expected to
Brogaard et al. (2013) show that HFT trades are more price increase HFT/AT and therefore CTR. The pre-MiFID mean
discovery efficient by virtue of trading in the direction of CTR rose from 3.07 to 17.91 post-MiFID in London (see
permanent price moves rather than in the direction of Table 1, Panel A) and from 1.42 to 8.01 in Paris (see
transitory shocks. Table 1, Panel B), both significant at 1 %. Cutting the data
into end-of-quarter (EOQ) expiry days (when some fairness
violations are more numerous) in Panels C and D, CTR is
Data again much higher post-MiFID1 introduction (3.82–17.57
in London and 1.33–7.58 in Paris). Subsequently, Exhibit 2
For all listed securities on the LSE and Euronext, we shows that once LSE held discussion boards of inquiry
retrieve the bid and ask quotes (time stamped to the nearest about a sliding scale cost recovery fee (finally introduced in
millisecond) for up to ten levels of the limit order book November 2010), CTR returned to its pre-MiFID1 levels
from 2003 through 2011. The data come from Thompson but thereafter continued its ascent.
Reuters Tick History (TRTH) and to construct our trading-
ahead measure of market fairness, we also use Reuters 12
NASDAQ designates access ports as HFT/AT based on the CTR,
News Announcement (RNA) feed. We eliminate reduc-
the order-to-trade ratio, zero inventory positions held overnight, speed
tions in depth due to trade executions and classify the attributable to co-location and stated intent. We do not employ as a
proxy for HFT/AT the OTR adopted by Hendershott et al. (2011).
Although it is tempting to suggest CTR might be an instrument for
11
The referee cautions quite appropriately that correlation is not identifying OTR, the covariance between CTR and residual distur-
causation in these studies. Later in the paper, we will perform bances at the individual stock level such as state of the market regime
exogeneity, strong instruments and other specification tests to address shocks is surely non-zero. Hence, CTR would be endogenous and
this issue. therefore invalid as an instrument.
123
45
40
LSE Euronext-Paris
35
30
Euronext-Paris
25
20
15 MiFID1
Nov 2007
10 LSE
0
1/2/2001 1/2/2002 1/2/2003 1/2/2004 1/2/2005 1/2/ 2006 1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 French
LSE’s Sliding Scale Trans
Cost Recovery Fee Tax
November 2010 Aug 2012
Exhibit 2 Cancel-to-trade (CTR) ratio, a proxy for algorithmic trading
Market Quality Metrics for Fairness violation alert based on surveillance industry practices for
and Efficiency identifying possible insider trading. Detection leads to
investigation, discovery, indictment, settlement negotia-
The evidence to date indicates that increasing HFT/AT tion, prosecution and rarely a conviction. Detection alone
results in increased efficiency. Ideally, the increase in using the LEAKAGE metric therefore provides an upper
efficiency should occur without harming the fairness of the bound on insider trading in a marketplace.
market. Indeed, this is one of the main criticisms raised by Measurement of LEAKAGE begins by identifying unu-
market participants concerning the impact of HFT/AT. To sual price and volume behaviour prior to price-sensitive
assess this issue, we develop two fairness metrics, as announcements. First, we calculate: AbnormalReturnit ¼
detailed below. Returnit bi ðReturnmt Þwhere Returni is the daily return on
stock i, bi is the stock-specific correlation with the market
Market Fairness—Information Leakage return constructed over the benchmarking period and Re-
turnm is the daily market return. Our metric is based upon
LEAKAGE measures the dissemination of private price- price-sensitive company announcements, where an
sensitive information ahead of a public announcement. announcement is regarded as price-sensitive when the
When human informed participants trade on their informa- return of the underlying security between days t - 6 and
tion, they create order imbalance that HFT/ATs can then t ? 2 is more than three standard deviations (3ri) away
identify and trade on themselves. A second reason HFT/AT from the average seven-day abnormal return constructed
may be correlated with LEAKAGE is related to the ability of using a bootstrap procedure during a 250 trading day
mechanical processes to almost instantaneously incorporate benchmarking period ending at t - 10. For each trading
changes in publicly disclosed information. With information day, the bootstrap procedure computes 1000 9-day returns,
providers such as Reuters and Bloomberg now selling real- with replacement.
time feeds in a format that is friendly to mechanical analysis, In addition to a market model of returns, our metric uses
HFTs may simply be able to impound the impact of new a clean event window (similar to the FSA’s occasional
information more quickly than non-HFTs. paper 25 Measuring Market Cleanliness (Dubow and
The true level of insider trading is not directly observ- Monteiro 2006) that lacks any other stock-specific
able. Instead, our LEAKAGE metric provides a fairness announcements over a 6-day pre-event period. If a
123
Table 1 Descriptive statistics Table 1 continued

Panel A: LSE Panel B: Euronext End-of-quarter (EOQ) Panel C: LSE Panel D: Euronext
Mean SD Mean SD Mean SD Mean SD
MTC (marking the close) Post-MiFID1 0.01 0.06 -0.01 0.06

Pre-MiFID1 31.51 22.11 39.34** 14.34 Vlum
Post-MiFID1 7.55 4.62 31.81** 18.72 Pre-MiFID1 25 0.16 22.31 0.2
LEAKAGE Post-MiFID1 24.8 0.15 22.18 0.22
Pre-MiFID1 0.245 0.315 0.254 0.235 Std
Post-MiFID1 0.207 0.135 0.288 0.180 Pre-MiFID -3.12** 0.26 -2.94** 0.41
SPR (effective spread) Post-MiFID1 -2.48** 0.34 -2.28** 0.35
Pre-MiFID1 17.43 6.07 11.71 3.12 Price
Post-MiFID1 10.82 2.11 7.68 1.71 Pre-MiFID1 5.66 0.12 4.35* 0.09
AT (CTR, cancel-to-trade ratio) Post-MiFID1 5.61 0.12 4.03* 0.31
Pre-MiFID1 3.07 1.39 1.42 0.32
We present descriptive statistics for each of our variables. MTC
Post-MiFID1 17.91 11.13 8.01 6.28
(marking the close) are instances of end-of-day prices that are dis-
OTR (order-to-trade ratio) located by trade-based manipulation. The price change during the last
Pre-MiFID1 11.54 1.25 12.75 1.35 15 min is compared with its historical distribution over the previous
Post-MiFID1 56.03
1.53 61.11 1.47 30 trading days to identify cases that are outliers of more than 3
standard deviations. Then from this sample a subset that has a price
Ret reversion of 50 % or more in the first 15 min the next trading day are
Pre-MiFID1 -3.91 0.94 -3.60 1.00 selected. LEAKAGE is the ratio of the number of information leak-
Post-MiFID1 -3.49 1.01 -3.26 0.97 ages to the number of clean event windows 9 100. An event is the
occurrence of an information announcement, an event window com-
Vlum
prises the days t - 6 through t ? 2 around the announcement, and a
Pre-MiFID1 24.87 0.42 22.09 0.42 clean event window is an event window with no other information
Post-MiFID1 24.79 0.22 22.20 0.22 announcements. Abnormal returns are calculated for each
Std announcement as the difference between the individual stock return
for an event window and the return on the market index. We retain a
Pre-MiFID -2.85 0.51 -2.63 0.55 sample of price-sensitive announcements, which are announcements
Post-MiFID1 -2.45 0.48 -2.24 0.45 for which the abnormal return is more than 3 standard deviations from
Price the mean abnormal return for the 250-day base period ending at
t - 10. SPREAD, relative effective spread, is 200 times the trade
Pre-MiFID1 5.60 0.16 4.34 0.11
price times the trade direction indicator (1 for buys and -1 for sells)
Post-MiFID1 5.61 0.11 4.02 0.28 times the difference between the trade price and the quote midpoint
End-of-quarter (EOQ) Panel C: LSE Panel D: Euronext relative to the quote midpoint in basis points. Our proxy for high-
frequency trading, AT, is the ratio of the number of cancellations to
Mean SD Mean SD the number of trades (CTR). After eliminating reductions in depth due
to trade executions, we classify the remaining quote updates that only
MTC (marking the close) reduce depth as cancellations. OTR is the L10 order book’s order-to-
Pre-MiFID1 41.17 24.34 41.83** 11.21 trade ratio, a metric of order instruction message traffic. Ret is the
natural log of the mean index return for each market. Volatility is the
Post-MiFID1 7.88 5.77 31.18** 12.58
natural log of the mean standard deviation of daily index returns.
LEAKAGE Volume is the natural log of the total volume traded on all listed
Pre-MiFID1 0.250 0.131 0.203* 0.192 securities in each month. Pre-MiFID1 is for the period prior to
Post-MiFID1 0.221 0.144 0.252* 0.164 November, 2007 and Post-MiFID1 is for the subsequent period.
Quarter is a dummy variable to account for the end of each financial
SPR (effective spread) quarter. EOQ refers to end-of-quarter data. For each variable for each
Pre-MiFID1 13.74 1.68 9.94* 1.93 Panel, we test the null hypothesis of equality of means, Pre and Post-
Post-MiFID1 10.83 1.88 7.47* 1.48 MiFID1 using a Wilcoxon rank-sum difference in means test

AT (CTR, cancel-to-trade ratio) and ** indicates significance at 1 and 5 %, respectively

Pre-MiFID1 3.82 0.9 1.33 0.25
Post-MiFID1 17.57 11.58 7.58 6.3
company has more than one announcement, only the first
announcement is used. Information leakage is proxied as an
OTR (order-to-trade ratio)
abnormal price movement on one or more days in the t - 6
Pre-MiFID1 11.70 2.78 12.68 4.47
to t - 1 pre-announcement period in the same direction as
Post-MiFID1 60.53 24.27 64.23 24.48
the overall announcement return. For a positive
Ret
announcement, information leakage is defined as any
Pre-MiFID1 0.01 0.02 0.02 0.03
1
P260
abnormal return [ ARi þ 3 ri where ARi ¼ 250 t¼10
123
Abnormal returnit ; where day t occurs between event day that traders are mimicking one another in order to ramp
-6 and -1, and where there are no days in which the market.
Abnormal returnit \ARi 3 ri for any day in the pre- We estimate suspected instances of ramping manipula-
event window. LEAKAGE is then defined as the ratio of tion at the close also using surveillance industry alert
information leakage events to clean event windows. procedures. The surveillance industry devotes much time
and effort and has developed considerable expertise in
Market Fairness—Market Manipulation distinguishing purely abnormal closing prices from ramp-
ing manipulations marking the close (MTC). MTC is sus-
Market manipulation involves creating a false or mislead- pected when an EOD percentage price change (in the last
ing representation with the intent to dislocate the market 15 min of trading plus the closing auction) Rit exceeds 3ri
price. Determining an accurate closing price is critical for above or below the mean of 30 days prior observations and
many market participants because of the multifaceted role then mean reverts at least 50 % in the first 15 min of
of marking-to-market in many instruments and in reporting trading the next day. Statistically abnormal EOD returns
many types of transactions. As such, market manipulation are then screened against Reuters’s database for textual
at the close reduces the information content of stock prices evidence of false rumours or company announcements. The
and thereby degrades market quality. remainder is defined as successful ramping manipulation
There are many types of market manipulation involving attempts to mark the close (MTC).
partial and misleading disclosures or false rumours but our As many (but not all) of the motivations for EOD
research focuses on trade-based manipulation at the close manipulation relate to reporting periods, we would expect
(defined to include any closing auction). Potential moti- MTC to be higher at the end-of-quarters. In Table 1, the
vations for ramping closing prices include: first two rows of Panels A and C show that on the LSE pre-
• To modify the value of managed funds at the end of MiFID1, this is clearly the case. Mean incidence of MTC in
reporting periods; EOQ months across all LSE-listed stocks is 41.17 events.
• To profit from derivatives positions in the underlying From the 8 years of data and the universal mean MTC of
stock; 31.85 events, we can calculate the mean in non-EOQ
• To obtain a favourable price in pre-arranged off-market months as 28.74 events, statistically significantly lower at
trades; 1 % in a non-parametric rank-sum difference test. Post-
• To alter their customers’ inference of broker execution MiFID1, the London MTC data from EOQ and non-EOQ
ability; months do not differ. Interestingly, MTC incidence for
• To maintain a stock’s listing on an exchange with Euronext-Paris (in Table 1, Panels B, D) is very nearly the
minimum price requirements; same both before and after MiFID1. We address later the
• To gain inclusion in an index near stock index multivariate, partial effect of EOQ which proves to be
rebalancing days and asymmetric around MiFID1.
• To avoid margin calls.
Market Efficiency—Execution Costs
Clearly, not all abnormal end-of-day (EOD) prices are
the result of trading strategies intended to dislocate the The execution cost to trade various sizes is measured by the
market close, and therefore not all trading behaviour that volume-weighted relative effective spread on a per trade
appears associated with run-ups or collapsing price at the basis and then averaged across the day. We calculate relative
close is illegal market manipulation. Some reasons why effective spreads in the standard fashion as the trade price
stock prices may naturally close at levels that appear sta- minus the midpoint of the bid-ask spread immediately prior
tistically abnormal from the perspective of the rest of the to trades. Daily and monthly spreads are equally weighted
trading day include: averages of all the securities traded that day or month.
• Some market participants may refuse to hold inventory
overnight and instead liquidate at the close;
• Brokers with a mandate to establish certain stock Research Design
positions or who guarantee Value-Weighted Average
Price (VWAP) at the close each day to their investors A Simultaneous Equations Model for Market
may be forced to become aggressive and are often Quality Research
‘mousetrapped’ by other trading desks at the EOD;
• Positive serially correlated trading patterns as HFT/ATs Market fairness, spreads and AT are simultaneously
seek to go flat at the EOD may create a false impression determined. An increase in market manipulation (MTC in
123
equation i below) raises price volatility which reduces minimises the Akaike information criterion proved to be 1
order aggressiveness, leading to higher effective spreads. or 2 lags for the LSE and for Euronext-Paris. These lag
Ceteris paribus, higher spreads (SPR in equation ii) reduce structures correct for the autocorrelation present in the
the incidence of manipulation, especially in more thinly monthly level data. Pooling tests fail across our two
traded stocks, because of the higher costs of trading. exchanges but allow aggregation of 2003–2011 trading
Finally, both fairness violations and trading costs will data before and after MiFID1. Each of the endogenous
influence the execution channel decision to employ AT. variables in simultaneous equation system (1)–(3) could in
And, given AT’s capabilities in assessing the real-time principle affect all the others, and consequently parameters
state of the market, an increase in AT may lead to a b1 and b2, b9 and b10 and b17 and b18 (which are pivotal to
reduced potential to manipulate the close, greater liquidity the fairness-efficiency framework that we are proposing)
and lower spreads. could exhibit substantial simultaneity bias. It is therefore
crucial to devote considerable attention to endogeneity
Fairness Eqn: MTC ¼ ð d c ; Control variables;
Spr; AT
tests and identifying restrictions. We fulfil the order con-
Market Design Changes; Fixed effectsÞ ðiÞ dition for identification by excluding from each equation at
d AT
Efficiency Eqn: SPRi ¼ ð MTC; c ; Control variables; least two exogenous instruments or control variables that
Market Design Changes; Fixed effectsÞ ðiiÞ are statistically significant elsewhere in the system. We
exclude (1) Price, OTR, Fee and Year from the MTC
d d
AT Eqn: ATi ¼ ð MTC; Spr; Control variables; equation, (2) OTR, Fee and EOQ from the QSpr equation
Market Design Changes; Fixed effectsÞ ðiiiÞ; and (3) Price, EOQ and Year from the AT equation. In an
online Technical Appendix, we analyse these over-identi-
where MTCi,t is the marking the close Fairness violation (or fying restrictions, and perform both Hausman and Stock–
alternatively Information Leakage, LEAKAGE), SPRi,t is Yogo pretests of the exogeneity and strength of our AT
the effective spread and ATi,t is the algorithmic trading instruments.
proxied by the cancel-to-trade ratio.
We specify the following regression relations estimated The Importance of Systems Estimation
over i,t stock-months 2003–2011:
d it þ b2 d
MTCit ¼ a þ b1 SPR ATt þ b3 Stdit þ b4 Retit For London and Paris, then, we estimate instrumental vari-
þ b5 Volit þ b6 AIit1 þ b7 MiFID þ b8 Quarter þ ~0 d t ; SPR
able models using MTC d t and AT c t instruments based on
full reduced-form equations of all the exogenous and pre-
ð1Þ
determined variables in (1)–(3). A priori, we wish to
d it þ b10 AT
SPRit ¼ a0 þ b9 MTC c it þ b11 Priceit þ b12 Volit emphasise the crucial importance of systems modelling of
þ b13 Std it þ b14 SPRit1 þ b15 MiFID þ b16 Year þ ~00 fairness-efficiency relationships. One could ignore the cross-
equation correlation of residual shocks to liquidity affecting
ð2Þ
the Spread, to the presence of an insider or market manipu-
d it þ b18 SPR
ATit ¼ a00 þ b17 MTC d it þ b19 Std it þ b20 Volit lator affecting MTC or LEAKAGE, or to the order cancel-
þ b21 MiFID þ b22 Retit þ b23 OTRit þ b24 Feeit þ ~000 ; lation proxy for AT activity, and then plough ahead to
estimate the full reduced-form instruments for
ð3Þ d t ; MTC;
d LEAKAGE;d c t based on all the exoge-
SPR and AT
where STDit is the mean standard deviation of daily returns, nous variables in the 3-equation system. Using these
Retit is the mean daily return, Volit is the mean turnover, instrumental variables in each equation separately yields the
MiFID is the Dummy variable for the introduction of two-stage least squares (2SLS) estimates. If the variance–
MiFID1 (=0 prior to November 2007 and 1 thereafter), covariance matrix of residuals from these 2SLS equations is
Quarter is the Dummy variable to account for the end of a diagonal, the 2SLS parameters will be consistent and
financial quarter, Priceit is the mean price, Year is the asymptotically efficient. In that case, three-stage generalised
yearly fixed effects, OTRit is the order-to-trade ratio, Feeit least squares estimates (3SLS) incorporating the off-diago-
is the dummy variable for transaction cost recovery fee (=0 nal residual variances and covariances will be identical to
prior to November 2010 and 1 thereafter) and ~0 ; ~00 ; ~000 are 2SLS estimates. The best indication this will not be the case
the residual error terms that may well be cross-equation here is the magnitude of the cross-equation correlation of the
correlated. residuals of estimated Eqs. (1)–(3). In London, these resid-
All variables were tested for persistence and found to be ual correlations run as high as 0.551 between AT and Spread,
stationary in the detrended monthly data. Depending on the and 0.790 between AT and Leakage. In Paris, the Spread and
functional specification, the optimal system lag length that Leakage residuals are correlated -0.698, and the MTC and
123
AT residuals are correlated 0.547. 2SLS estimates have been increased scrutiny at precisely these EOQ and expiry days
estimated, compared and found to differ in several note- could cause trading ahead of price-sensitive announcements
worthy ways from our 3SLS models. The fact that our results (LEAKAGE) to decline. However, it appears that the
are especially sensitive to the specification errors associated extraordinary dispersion of order flow that accompanied
with omitting the information contained in the variance– MiFID1 made the normal EOQ scrutiny much less effective.
covariance matrix of residuals highlights the need for care- Controlling for many confounding variables, we find in our
fully incorporating into market microstructure empirical multivariate simultaneous equations analysis (as sum-
models the entire market quality framework we have pro- marised in Exhibit 4) that in top decile stocks, MiFID1
posed. Market efficiency and market fairness interact; part of increased MTC in London and both MTC and LEAKAGE in
that interaction is behavioural, but part is correlated random Paris. Of course, correlation does not imply causation; post-
disturbances attributable to various types of (unobservable) MiFID1 regulators may have been keener to enforce some
trading strategies favoured by the market participants under securities laws rather than others.
study. Value-weighted SPR exhibits significant reductions at
99 % confidence in univariate analysis of the post-MiFID
period for both exchanges in both samples. Exhibit 3 on
Discussion of Results relative effective spreads in these European markets shows
spreads eroding from 20–35 to 10–15 b.p. earlier in the
Descriptive Statistics decade. Then when MiFID1 was introduced, value-
weighted spreads initially spiked and thereafter fell con-
For both the LSE and Euronext full samples and end-of- sistently throughout 2008–2010, by 40 % on the LSE and
quarter (EOQ) subsamples, Table 1 presents descriptive by 60 % on Euronext-Paris. This is precisely when an
statistics before and after the introduction of MiFID1. The increasing market share of the trading migrated to the Chi-
implementation of MiFID1 in November 2007, accompa- X multilateral trading system, a venue that proved espe-
nied by direct market access and co-location, allowed cially attractive to high-frequency traders.
algorithmic traders using an MTF such as Chi-X to avoid Concurrent increases in AT after 2007 led to other
all structural barriers to reaching the national exchanges. massive reductions in spreads. Why would effective
Posting and cancelling and revising quotes with dizzying spreads narrow with greater AT? For one thing, techno-
frequency, the Level 10 limit order message traffic explo- logical innovation substantially reduces order processing
ded thereafter, rising to 200 orders/trade by August 2010. costs that are normally recovered through the effective
Exhibit 2 shows that our empirical proxy for AT (the spread. In addition, because of AT’s distinctive capabilities
cancel-to-trade ratio) reached a high point in November in fast assessment and execution of alternative trading
2010 in London and in January 2011 in Paris. strategies triggered by changes in the state of the market,
Remembering that these univariate effects of introduc- adverse selection costs decline. Jovanovic and Menkveld
ing MiFID1 may well not hold up as ceteris paribus, partial (2011) predicted this fortuitous result. Even after
effects when we estimate the concurrent effect of AT in the accounting for simultaneity and many control variables (as
equation system (1)–(3), Table 1, Panel A shows MTC summarised in Exhibit 4), we find that AT reduced and
declined significantly in London following the introduction MiFID1 raised spreads in London-listed stocks.
of MiFID1 (compare mean events of 31.85 before to 7.55
after, significant at 99 %). Similarly in EOQ months (Panel 3SLS Systems Estimation for Liquid Stocks
C), MTC declined significantly following MiFID1, with
smaller but still statistically significant reductions at 99 % In our multivariate analysis, we examine first the stocks in
for Euronext in both the universal sample (Panel B) and the the top 2 most actively traded deciles, rebalanced annually,
EOQ subsample (Panel D). Highlighting the idea that MTC which gives a sample of approximately 570 stocks on the
represents induced (not natural) volatility, post-MiFID1 we LSE and 220 on Euronext-Paris. In Table 2, we present the
observe in EOQ months in both London and Paris (Panels results of our 3SLS estimation for these most liquid stocks
C, D) significant reductions at 99 % in natural price on LSE in Panel A and on Euronext-Paris in Panel B. The
volatility (Std) as well. first and fourth columns display alternate dependent vari-
On the other hand, the univariate impact of MiFID1 on ables (MTC and LEAKAGE) to proxy for the two different
LEAKAGE is in the opposite direction. MiFID1 increased types of market fairness violations. Higher AT in columns
information leakage in Euronext-Paris at the end-of-quarters 3 and 6 is inversely related to spreads and cost recovery
(see Table 1, Panel D) with 99 % confidence. Although we fees (Fee) in London, as hypothesised, and proves not to be
would expect trade-based manipulation to exhibit higher associated with high volatility (Std) in these most liquid
incidence at these reporting deadlines and expiry dates, stocks. In column 6, lower AT is also consistently
123
35
Relative Effective Spread 2001-2011
Effective Bid-Ask Spread (Basis Points) 30
Nov 2003 – Nov 2011

25
Sample period
20
Chi-X
15 Mkt Share↑
10
Spreads of 20-35 Euronext-Paris

5 bp erode to MiFID1
10-15bp by 2003 Nov 2007 < 10bp
0
2001-1
2001-4
2001-7
2001-10
2002-2
2002-5
2002-8
2002-11
2003-2
2003-5
2003-9
2003-12
2004-3
2004-6
2004-9
2004-12
2005-3
2005-6
2005-9
2005-12
2006-3
2006-7
2006-10
2007-1
2007-4
2007-7
2007-10
2008-1
2008-4
2008-7
2008-11
2009-2
2009-5
2009-8
2009-11
2010-2
2010-5
2010-8
2010-11
2011-2
2011-5
2011-8
2011-11
Exhibit 3 Effective spreads by month 2001–2011
As an explanatory variable, AT in row 7 is unrelated to

Effects of Increased AT MTC alerts in column 1 but is negatively and significantly
London Paris related to LEAKAGE alerts in column 4 for both exchan-
ges. So, AT improves what the U.K.’s Financial Service
All Stocks
Authority calls the ‘‘cleanliness’’ of the information-pro-
Effective Spreads ↓ ↓ ↓ Effective Spreads ↑ ↑ ↑
cessing environment. Furthermore, in addition to improv-
Leakage ↓ ↓ ↓ ing one of the two dimensions of market fairness, greater
Manipulation ↓ ↓ ↓ AT lowers the SPREAD in both fairness submodels (col-
_________________________________________________________ umns 2 and 4) and does so in both markets (Panels A, B).
Top Quintile Stocks Note that our simultaneous equations model makes clear
Effective Spreads ↓ Effective Spreads ↓
that fewer fairness violations offer a double benefit in that
LEAKAGE in London and Paris (row 3 of Panels A, B) is
Leakage ↓ ↓ ↓ Leakage ↓
_________________________________________________________ itself a significant positive determinant of wider spreads in
column 5. So, spreads decline both directly with greater AT
Effects of MiFID1 and indirectly because AT reduces the number of LEAK-
London Paris AGE fairness violations as a percentage of price-sensitive
All Stocks announcements.
Effective Spreads ↑ Effective Spreads ↓ In stark contrast, MiFID1 in row 8 is associated with
Leakage ↑ ↑ ↑ Leakage ↑ ↑ ↑ more market manipulation in both London and Paris (col-
Manipulation ↑ umn 1 of Panels A, B) and more information leakage in
___________________________________________________________
Paris (column 4 of Panel B). In discussions as recently as
Top Quintile Stocks
March 2012, the Economic and Monetary Affairs Com-
Leakage ↑ ↑ ↑
mittee rapporteur Arlene McCarthy argued that the order
Manipulation ↑ ↑ ↑ Manipulation ↑ flow dispersion which MiFID1 has fostered makes it more
difficult to detect instances of misconduct.13 In row 9,
Exhibit 4 Summary of multivariate results for AT and for MiFID1
reduced-form predicted Spread ( d Spr) is negatively related
to Manipulation (MTC) in London (column 1, Panel A) but
positively related to MTC in Paris (column 1, Panel B).
associated with the instrument for higher LEAKAGE
Similarly, the end-of-quarter (EOQ) dummy variable in
suggesting real-time models of the ‘‘state-of-the-market’’
and the follow-on automated trading are discouraged by the
13
presence of insiders who leak new valuation information. European Parliament News Release—12 April, 2012.
123
Table 2 Determinants of market fairness, effective spread and algorithmic trading for liquid stocks (3SLS estimation)
Top quintile stocks
Equation AI = Marking the close (MTC) AI = LEAKAGE
MTC SPREAD AT LEAKAGE SPREAD AT
a
Panel A: LSE
Intercept -79.005 -0.420* 56.581 0.721 -1.576 10.297
(-0.83) (-1.71) (3.99) (4.36) (-3.34) (0.38)
dt
MTC – -0.0004 -0.003 – – –
– (-1.03) (-0.13) – – –
d
LEAKAGE t
– – – – 0.481 -19.907
– – – – (4.91) (-8.47)
Price – -0.021 – – -0.011 –
– (-3.50) – – (-0.92) –
Vlum 3.323 0.023** -2.327 -0.019 0.059 -0.198
(0.85) (2.30) (-4.01) (-0.28) (3.27) (-0.02)
Std 0.353 0.011** 0.189 -0.008 -0.001 0.461
(0.15) (2.21) (0.62) (-0.02) (-0.10) (0.69)
ct
AT -0.003 -0.004 – -0.219 -0.109 –
(-0.11) (-5.70) – (-8.11) (-6.06) –
MiFID1 11.283** -0.022 -0.580 0.094 -0.011 0.356
(2.31) (-1.46) (-0.72) (1.32) (-0.41) (0.32)
dt
SPR -5.869* – -0.837* -0.067 – -1.278**
(-1.62) – (-1.69) (-1.29) – (-2.02)
Return -10.252 – -0.678 -0.315 – -4.270
(-0.59) – (-0.27) -(1.30) – (-1.09)
EOQ -0.185 – – -0.028 – –
(-0.11) – – (-2.50) – –
Fee – – -1.337 – – -0.438
– – (-2.71) – – (-1.12)
Durbin–Watson 2.05 2.03 2.05 1.93 1.87 1.97
TSLS adjusted R2 0.617 0.967 0.665 0.158 0.927 0.443
Year fixed effects No Yes No No Yes No
3SLS, top quintile stocks
b
Panel B: Euronext-Paris
Intercept -1.942 -0.217 -115.02 -0.438 -0.118 -27.037
(-0.86) (-1.47) (-2.82) (-0.18) (-0.74) (-0.65)
dt
MTC – 0.0001 -0.258 – – –
– (0.54) (-5.37) – – –
d
LEAKAGE t
– – – – 0.047** -20.52
– – – – (2.47) (-5.73)

Price – -0.034 – – -0.042 –
– (-4.85) – – (-5.25) –
Vlum 0.116 0.019** 5.473 0.051 0.018** 1.776
(1.28) (2.71) (2.95) (0.47) (2.57) (0.95)
123
Table 2 continued
3SLS, top quintile stocks

Std 0.005 0.003 2.452 -0.082 0.008** -0.898
(0.14) (1.03) (3.07) (-1.43) (2.67) (-0.96)
ct
AT -0.011 -0.009** – -0.097 -0.014 –
(-0.55) (-2.26) – (-3.03) (-3.50) –
MiFID1 0.096* -0.004 -2.859** 0.323 -0.005 4.874
(1.71) (-0.80) (-2.39) (4.14) (-0.52) (4.13)
dt
SPR 0.386 – -4.611 0.307 – -4.037**
(4.20) – (-3.13) (3.26) – (-2.52)
Return -0.118 – -33.674 -0.609 – -9.170
(-0.54) – (-3.84) (-1.44) – (-1.19)
EOQ -0.000 – – 0.077 – –
(-0.02) – – (3.34) – –
Durbin–Watson 1.73 1.69 1.82 1.89 1.76 1.86
TSLS adjusted R2 0.552 0.805 0.846 0.265 0.787 0.726
For stocks in the top 2 deciles by trading volume rebalanced each year for the period 2003–2011, let alert incidence AIt be either MTC or
LEAKAGE; SPRt = Effective Spread; ATt = CTR (Cancel-to-Trade Ratio). Using 3SLS we estimate the three-equation system (1)–(3) to
account for the cross-equation correlation of the ~0 ; ~00 ; ~000 We estimated the equations separately for the LSE and Euronext having tested and
found the pooled estimations invalid. T-scores with robust standard errors are reported in parentheses. Panel A reports results for the LSE and
Panel B for Euronext
a
T-scores with robust standard errors are in parentheses. Two lags are significant in four of these six equations; one lag is significant in both
Spread equations
b
T-scores with robust standard errors are in parentheses. A first lag is significant in all six of these equations

, ** and * indicates significance at 1, 5 and 10 %, respectively
row 11 is unrelated to Manipulation in either venue but The precursor to insider trading behaviour (information
negatively related to Leakage in London (column 4, Panel leakage prior to price-sensitive announcements) proves
A) whilst positively related to Leakage in Paris (column 4, much more difficult to model with adjusted R2 of only
Panel B). Differential commitment to enforcement could 0.158 in London (column 6, Panel A) and 0.265 in Paris
easily explain this difference. (column 6, Panel B). Few hypothesised determinants of
Our control variables behave as predicted, higher Leakage are statistically significant with anticipated signs.
Price is associated with significant reductions in the And recall that Stock–Yogo testing revealed that parameter
Spread (in basis points) across three of the four speci- bias may well be present because of the weakness of the
fications. Vlum is positively related to Spread as a Leakage instruments. As foreshadowed by these specifi-
reflection of more difficult trades. Where Std is found to cation pretests, we should therefore focus the preponder-
be significant, it increases Spread, as anticipated from ance of our interpretation and discussion on the MTC-
optimal order placement theory. Including fixed effects Spread-AT results.
for each year reveals deterministic downward trend in
spreads through our estimation period attributable to the All Listed Stocks
growth and entry of MTFs like Instinet-Europe, Chi-X
and BATS. These annual fixed effects in columns 2 and Employing all listed stocks from all liquidity deciles, 3SLS
5 drive adjusted R2 in the TSLS estimations to 0.967 and results for the MTC-Spread-AT submodel are reported on
0.927 in the two submodels for London and to 0.805 and the left-hand side of Table 3, and similarly for the Leak-
0.787 in the two submodels for Paris. The variation in age-Spread-AT submodel on the right-hand side. Just as in
manipulation at the close (MTC) is well modelled with the most liquid stocks, AT in columns 3 and 6 is inversely
adjusted R2 of 0.617 and 0.552 in London and Paris, related to cost recovery fees (Fee), predicted spreads ( d
Spr)
respectively. and Leakage fairness violations whilst being positively
123
Table 3 Determinants of market fairness, effective spread and algorithmic trading (3SLS—all listed stocks)
Panel A: LSE
Intercept -522.3* 0.514 55.33 -2.347 0.426** 156.11
(2.17) (2.95) (1.19) (-1.08) (1.95) (1.22)

dt
MTC – 0.0003 0.003 – – –
– (2.94) (0.14) – – –
d
LEAKAGE t
– – – – 0.114 -29.671
– – – – (8.14) (-3.15)
Price – -0.075 – – -0.089 –
– (-5.35) – – (-6.35) –
Vlum 21.39** 0.003 -2.403 0.130 0.006 -6.879
(2.27) (0.49) (-1.28) (1.45) (0.76) (-1.31)
Std 0.277 0.009 1.385** -0.008 -0.002 4.688**
(0.17) (4.50) (1.93) (-0.36) (-0.68) (2.36)

ct
AT -1.236 -0.019 – -0.024 -0.010** –
(-0.51) (-4.75) – (-1.20) (-1.98) –
MiFID1 -2.272 0.026 -1.096 0.104* 0.005 5.019*
(-0.34) (3.71) (-0.88) (1.83) (0.63) (1.79)
dt
SPR 0.481 – -5.288** 0.252 – -20.773
(0.48) – (-2.46) (2.86) – (-3.99)
Return -16.096 – -0.540 -0.109 – 16.460
(-0.49) – (0.09) (-0.50) – (0.88)
EOQ 2.567 – – 0.001 – –
(1.03) – – (0.06) – –
Fee – – -1.957 – – -4.018*
– – (-2.68) – – (-1.89)
Durbin–Watson 1.67 1.76 2.05 1.73 1.58 2.00
TSLS adjusted R2 0.740 0.938 0.953 0.188 0.966 0.585
3SLS, all listed stocks
Panel B: Euronext-Paris
Intercept -69.85** 7.181* -26.41** -3.139 12.156 -24.040
(-0.94) (1.82) (-2.10) (-1.47) (2.22) (-1.58)
dt
MTC – 0.020 -0.184 – – –
– (2.86) (-5.75) – – –
d
LEAKAGE t
– – – – 2.814 -8.482
– – – – (4.89) (-5.34)
Price – 0.005 – – -0.050 –
– (0.02) – – (-0.27) –
Vlum 9.311 -0.340** 1.529 0.267 -0.567** 1.341**
(2.79) (1.97) (2.64) (0.86) (-2.34) (1.95)
Std 5.66 -0.153 1.151 0.047 -0.016 0.545*
(3.16) (1.61) (3.79) (0.96) (-0.12) (1.67)
123
Table 3 continued
3SLS, all listed stocks

ct
AT -6.70 0.383 – -0.217 0.484 –
(-4.49) (3.57) – (-5.56) (4.03) –
MiFID1 7.138 -0.247 0.928 0.415 -1.210 -2.582
(2.64) (-1.07) (2.67) (5.46) (-4.41) (-5.97)
dt
SPR 7.263 – 1.221 0.227 – 1.221
(4.87) – (4.75) (7.32) – (4.75)
Ret -47.431 – -7.026** 0.098 – 2.895
(3.09) – (2.14) (0.49) – (1.35)
EOQ 1.502 – – -0.024 – –
(1.48) – – (-1.49) – –
Durbin–Watson 1.79 1.80 1.93 1.72 1.68 1.74
TSLS adjusted R2 0.415 0.372 0.824 0.163 0.685 0.747
Using 3SLS, we again estimate the three-equation system in Table 2 for all listed stocks first on the London Stock Exchange and then on
Euronext-Paris

, **,* indicates significance at 1, 5 and 10 %, respectively. T-scores with robust standard errors are in parentheses. Two lags are significant in
three of these six equations; one lag is significant in the Spreads and third column AT equations
related anew to volatility (Std). In Paris, AT is inversely 7), in London, Panel A shows that (just as in the most liquid
related to both LEAKAGE and MTC and again positively stocks) greater AT reduces effective spreads (in columns 2
related to volatility and volume of trading, as in the most and 5) but raises effective spreads in Paris (columns 2 and 5,
liquid stocks. Opposite to London, in Paris higher effective Panel B). Unlike in the top quintile stocks, however, there is
Spread enhances AT perhaps by triggering rent-seeking no significant effect of AT reducing manipulation (MTC) in
competitive trading across MTFs. Finally, MiFID1 is London (column 1, Panel A) and a large negative effect in
positive and significant as a determinant of AT in the MTC Paris (columns 1, Panel B). As to LEAKAGE in column 4,
submodel and negative and significant in the LEAKAGE unlike in the top quintile stocks, there is no significant effect
submodel. Dispersion of the order flow facilitated manip- of AT in London (column 4, Panel A) and like in the top
ulation in thinner venues but made surveillance and the quintile stocks, a large negative effect in Paris (columns 4,
detection of insider trading much more effective. Panel B). Nevertheless, having checked for both fairness
Turning next to the determinants of effective spreads (in effects, one can say unambiguously in our MQ framework
columns 2 and 5 of Panels A, B), we find spreads are pos- that AT raises market fairness in all listed stocks in both
itively and significantly related to the number of fairness markets. However, Panel B columns 2 and 5 show that AT
violations in both submodels in both London and Paris. This raised effective spreads in Paris.
evidence indicates strong support for our central hypothesis Hence, we are presented with market quality trade-offs.
that trade-based manipulation (in row 2) and information Our framework makes possible the estimation of these
LEAKAGE (in row 3) increase execution costs. With magnitudes for assessing the competing effects on market
heightened perceptions of dislocation of the price or insider quality. AT reduced fairness violations of both types in
trading, limit order traders and/or market makers back away Paris but raised the execution cost of completing trades.
from their inside quotes and price protect against picking The magnitude of reductions in the effective spread are
off risk. Furthermore, we know trade-based manipulation much larger for these less liquid stocks than for the top 2
increases price volatility thereby reducing order placement decile stocks. The Parisian regulators and exchange oper-
aggressiveness as a result of reduced non-execution risk. ators had to decide whether a transaction tax on AT was
As usual in microstructure models, effective spreads are warranted based on the higher spreads but lower fairness
lower with increases in the control variable Price (in row 4). violations attributable to greater AT. The introduction of a
Volume (in row 5) decreases and volatility (Std in row 6) stiff tax on message traffic in August 2012 suggests that
increases effective spreads in these less liquid stocks, as what we estimate as the large size of the fairness gains may
expected. As to the effects of the instrument for AT (in row not have been given sufficient account in the decision.
123
Table 4 Strong instruments in 3SLS estimation of marking the close, effective spread and algorithmic trading in London and Paris (all listed
stocks)
AI = Marking the close (MTC)
LSE Euronext
MTC SPREAD AT MTC SPREAD AT
Intercept -19.467** 14.397 -0.506 -10.853 82.546 -1.251

(-2.17) (4.85) (-0.11) (-3.07) (3.14) (-0.24)
d
MTC – 0.009 -0.105 – 0.062 -0.439
– (0.34) (-1.06) – (0.91) (-1.22)
d
LEAKAGE – – – – – –
– – – – – –
OTR – – 0.216** – – 0.655
– – (2.01) – – (5.60)
Price – -0.202** – – -0.373 –
– (-2.19) – – (-4.29) –

Vlum 0.880** 0.114 0.035 0.616 0.258 0.062
(2.42) (2.51) (0.19) (3.95) (3.36) (0.22)
Std 0.161 0.079 0.197* 0.179** 0.045 0.060
(0.97) (4.44) (1.96) (2.21) (0.136) (0.58)
ct
AT -0.382 -0.080 – -0.182 0.009 –
(-3.12) (-5.63) – (-2.53) (0.22) –
MiFID1 -0.610** 0.107* -0.054 -0.258* -0.051 -0.131
(-2.15) (1.92) (-0.32) (-0.190) (-0.92) (-0.86)
dt
SPR -0.611 – -0.570 -0.551 – -0.062
(-1.17) – (-1.33) (-3.00) – (-0.21)
Return 0.792 – 1.340 -3.625 – -0.926
(0.61) – (0.25) (-5.31) – (-0.63)
EOQ 0.075 – – 0.092 – –
(0.80) – – (1.57) – –
Fee – – -0.667 – – –
– – (-8.73) – – –
Durbin–Watson 1.94 1.91 2.04 1.96 1.91 2.07
TSLS adjusted R2 0.771 0.928 0.933 0.564 0.810 0.926
We re-estimate the three equations for Manipulation (MTC), Spread (SPR) and AT using the order-to-trade ratio OTR, a measure of message
traffic, as an instrument for AT. All continuous variables are in the natural logs

, ** and * indicates significance at 1, 5 and 10 %, respectively. T-scores with robust standard errors are in parentheses. One lag is significant in
four of these six equations. The MTC equations have two significant lags
MiFID1 (in row 8) in all listed stocks in London (Panel Outcomes-based policy decisions can now hinge on these
A) unambiguously lowers market quality because it magnitudes. Also, the fact that MiFID1 raised Spreads in
increases Spread (in column 2) without significantly London (column 2, Panel A) whilst lowering Spreads in
affecting MTC (in column 1) and yet increases Leakage (in Paris (column 5, Panel B) is a further indication that the
column 4). So, again using our framework, one can assert study of market quality must account for institutional dif-
MiFID1 doubly worsened the quality of the London mar- ferences that can prove pivotal.
ket. In Paris (Panel B), MiFID1 in all listed stocks lowers
effective spreads (in column 5) but raises both MTC (in Strong Instruments
column 1) and LEAKAGE (in column 4). Our estimates for
MiFID1 uncover an empirically quite large magnitude In Table 4, we re-estimate the MTC-SPR-AT submodel for
increase in both MTC and information Leakage offset by a all listed stocks in London and Paris using the order-to-
sizeable reduction in the Spread at least in Paris. trade ratio (OTR) as a strong and valid instrument for CTR,
123
the endogenous proxy metric of AT. Comparing Tables 3 propose less market integration in MiFID2 based on their
and 4, AT in columns 3 and 6 is again inversely related to combination of higher spreads and lower fairness violations
LSE’s sliding scale cost recovery fee, whilst being posi- accompanying MiFID1.
tively related to volatility (Std) and message traffic (OTR).
Only the last result holds in Paris where the French
transaction tax postdates our analysis. Again comparing Summary and Conclusion
Tables 3 and 4, MTC in columns 1 and 4 is in both
tables positively related to volume and volatility and Trade-to-trade data for 2003–2011 from the LSE and
inversely related to Return. NYSE-Euronext-Paris exhibit a massive increase in high-
Focusing on the first of our simultaneous variables, the frequency/AT following the 2007 introduction of the
d in Euronext is inversely related now in
predicted Spread MiFID1. We examine the impact of HFT/AT on the market
Table 4, column 4 with strong instruments to Manipulation quality of these two exchanges. Unlike previous studies
(MTC), as expected. In Table 4, Panel B, column without that have focused exclusively on market efficiency and
d in Euronext had been positively liquidity as measured by variables such as effective spreads
strong instruments, Spread
(SPR), price impact and volatility, we also examine the
related to MTC, inconsistent with the hypothesised theory
impact on market fairness. We construct two fairness
that successful ramping manipulation requires lower
proxies–-one for information leakage (LEAKAGE) and the
spreads especially in thinner stocks where execution costs
other for end-of-day manipulation—i.e. MTC. We then
can be quite high. So, here strong instruments made all the
develop and estimate a three-stage simultaneous equations
difference, and the 2SLS parameter exhibits a magnitude of
system that accounts for the endogenous relationships that
bias that proves material.
exist between AT, SPREAD and LEAKAGE or MTC, whilst
Comparing the determinants of spreads in Tables 3 and
incorporating information about unobservables embedded
4, Spread is in both tables positively related to fairness
in the cross-equation correlation of disturbances.14
d volume (Vlum) and volatility (Std) and
violations ( MTC), We document a significant change in both market effi-
inversely related to Price. As to the policy variables AT ciency and fairness on the LSE and Euronext-Paris, espe-
and MiFID1, AT lowers manipulation in both London cially apparent after the introduction of MiFID1. The
(Table 4, column 1) and Paris (Table 4, column 4) whilst policy implications of increased AT and the MiFID1
reducing Spread as well in London. These AT results just market design change are summarised in Exhibit 4. Our
parallel those in Table 4 without strong instruments. multivariate results indicate that the increasing level of AT
However, the effect of strong instruments on MiFID1 unambiguously increased market efficiency by reducing
results is indeed noteworthy. In Table 3, Panel A without relative effective spreads in London, and in top quintile
strong instruments, MiFID1 raises Spread in London stocks in both markets. This reduction in SPREAD is
(column 2) with no significant effect in Paris (column 4), consistent with theory that suggests because algorithmic
and we estimate the same result in Table 4, columns 2 and traders are better able to update their real-time assessments
4 with strong instruments. However, quite the opposite of the state of the market, they incur lower adverse selec-
happens when we examine the effect of MiFID1 on MTC. tion and inventory costs relative to other market
Without strong instruments (in Table 3), MiFID1 has no participants.
effect on MTC in London (Panel A, column 1) and raises Distinctively, we provide the first estimates on the
MTC in Paris (Panel B, column 1). With strong instru- relationship between AT and two metrics of market fair-
ments, the reverse occurs. MiFID1 reduces MTC in Lon- ness. We show the closing price is less manipulated the
don (column 1) and reverses the positive effect to a higher the incidence of AT in both London and Paris. By
negative in Paris (column 4). providing additional liquidity and reducing price impact,
Hence, with strong instruments we are again presented increased AT appears to raise the cost of implementing
with a market quality trade-off signified by the shadowed ramping manipulations. Moreover, we conjecture that like
oval in Table 4. MiFID1 reduced manipulation violations some hedge fund prop desks, ATs can reverse positions to
in London (perhaps by facilitating the arb trades of MTFs) profit from ramping manipulation events faster than the
but it raised effective spreads. The impact of MiFID1’s manipulators themselves, thereby removing the manipula-
widening London spreads (?0.107) is twice as large as the tor’s profit.
indirect effect tightening London spreads as a result of the
reduced manipulation (-0.610 9 0.009 = -0.055).
14
Again, our framework makes possible the estimation of the Our research highlights the interrelationships between market
magnitude of these competing effects. The London regu- fairness and transaction cost efficiency. The referee points out
correctly that additional insight in future research can come from
lators and exchange operators might well have decided to exploring the simultaneous effects on informational efficiency.
123
Our results regarding information leakage are strongest Aitken, M. J., & LeDuc, A. (2012). Fostering fair and efficient
in the most liquid stocks. Increased AT lowers the inci- securities markets—The primary objective of securities markets
regulators. Working Paper, CMCRC.
dence of leaking non-public information prior to price- Aitken, M. J., Harris, F. H. de B., McInish, T. H., Aspiris, A, & Foley,
sensitive announcements in the top 2 liquidity deciles on S. (2012). High frequency trading—Assessing the impact on
both Euronext-Paris and on the LSE. We conjecture that market efficiency and integrity. Foresight Project. The Future of
the speed and intraday forecasting accuracy of ATs in Computer Trading in Financial Markets, U.K. Government
Office for Science, Driver Review DR28.
separating permanent and transitory components makes Brogaard, J. (2010). High frequency trading and its impact on market
more abrupt the price reaction to trading based on infor- quality, Working Paper, Northwestern.
mation leakage. This more conspicuous footprint would Brogaard, J. A., Henderschott, T., & Riordan, R. (2013). High
markedly increase the probability of surveillance detection, frequency trading and price discovery. Review of Financial
Studies, 27(8), 2267–2306.
prosecution and conviction for insider trading, thereby Chakravarty, S., Harris, F. H. de B., & Wood, R. A. (2009). The
discouraging information leakage and trading ahead of changing state of the NYSE Market: New common factors,
price-sensitive announcements. players, and impulse responses. Journal of Trading, 4, 68–94.
The widely dispersed trading facilitated by MiFID1 Cvitanic, J., & Kirilenko, A. (2010). High frequency traders and asset
prices. SSRN eLibrary.
proved concurrent with increased market manipulation in Dubow, B., & Monteiro, N. (2006). Measuring market cleanliness.
London and Paris for both liquid and illiquid stocks. In Financial Services Authority Occasional Paper 23.
addition, Leakage increased in Paris, and spreads rose in Dugast, J., & Foucault, T. (2014). False news, information efficiency,
London after the introduction of MiFID1. One insight and price reversals, Working Paper, HEC.
Gomber, P., Arndt, B., Lutat, M., & Uhle, T. (2011). High-frequency
revealed by our research, therefore, is that AT saved the trading. SSRN eLibrary.
London and Paris markets from the massive market quality Harris, F. H. de B., & DiMarco, E. (2012). European market quality
reductions that would otherwise have occurred. Although pre and post MiFID. Journal of Trading, 7, 7–26.
AT was surely evolving on its own, it is fortunate indeed Harris, F. H. de B., McInish, T. H., Sensenbrenner, F., & Wood, R. A.
(2014). Fragmentation and price discovery: A comparison of
that direct market access allowed AT to explode soon after RegNMS and MiFID1. Journal of Trading, 9, 6–36.
MiFID1 was introduced. Harris, F. H. de B., McInish, T. H., & Wood, R. A. (2002). The
Overall, our new evidence dispels concerns that dynamics of price adjustment across exchanges: An investigation
increased AT would undermine market quality. Our findings of price discovery for Dow Stocks. Journal of Financial
Markets, 5, 277–308.
clearly indicate that not only did greater AT 2003–2011 Hasbrouck, J., & Saar, G. (2013). Low-latency trading. Journal of
reduce effective spreads in all stocks in London and in top Financial Markets, 16, 646–679.
quintile stocks in Paris but AT did so without harming the Hendershott, T., Jones, C., & Menkveld, A. (2011). Does algorithmic
fairness of either market. Indeed, manipulation at the close trading improve liquidity? Journal of Finance, 66, 1–33.
Hendershott, T., & Moulton, P. (2011). Automation, speed, and stock
and information leakage in top quintile stocks also declined market quality: The NYSE’s hybrid. Journal of Financial
in both venues with greater AT. We conclude that in Europe, Markets, 14, 568–604.
AT has enhanced market efficiency and market fairness Hendershott, T., & Riordan, R. (2013). Algorithmic trading and the
despite the now widely dispersed order flow. market for liquidity. Journal of Financial and Quantitative
Analysis, 48(4), 1001–1024.
Jarnecic, E., & Snape, M. (2010). An analysis of trades by high
Acknowledgments The U.K. Treasury sponsored this research frequency participants on the London Stock Exchange. Working
through their Foresight Project, The Future of Computer Trading in Paper, CMCRC.
Financial Markets: An International Perspective, Government Office Jovanovic, B., & Menkveld, A. J. (2011). Middlemen in limit-order
for Science, 2012. Aitken et al. (2012) provided our preliminary markets. SSRN eLibrary.
report. We wish to thank several Office of Science referees, seminar Kumar, P., Goldstein, M., Graves, F., & Borucki, L. (2011). Trading
participants at the U.S. Securities Exchange Commission, the Aus- at the speed of light: The impact of high-frequency trading on
tralian Securities and Investments Commission, the Italian CONSOB, market performance, regulatory oversight, and securities litiga-
and the Six Swiss Exchange, as well as the CMCRC development tion in Finance-Current Topics in Corporate Finance and
team and SIRCA for providing the raw data. An anonymous referee Securities Litigation, The Brattle Group, Issue 02/2011, pp. 1–12
provided exceptionally insightful perspectives. Kyle, A., & Vishwanathan, S. (2008). How to define illegal price
manipulation. American Economic Review, 98(2), 274–279.
McInish, T., & Upson, J. (2013). The quote exception rule: Giving
References high frequency traders: an unintended advantage. Financial
Management, 42, 481–501.
Aitken, M. J., & de B. Harris, F. H. (2011). Evidence-based policy Menkveld, A. J. (2013). High frequency trading and the new market
making for financial markets: A fairness and efficiency frame- makers. Journal of Financial Markets, 16, 712–740.
work for assessing market quality. The Journal of Trading, 6, Yan, B., & Zivot, E. (2010). A structural analysis of price discovery
22–31. measures. Journal of Financial Markets, 13, 1–19.
Aitken, M. J., Harris F. H. de B. & Ji, S. (2014), A worldwide Zhang, F. X. (2010). High frequency trading, stock volatility, and
examination of exchange market quality: Greater integrity price discovery. Working Paper, Yale University.
increases market efficiency. Journal of Business Ethics (pub-
lished Online, August 12). doi:10.1086/599247.
123

Market Fairness: The Poor Country Cousin of Market Efficiency

Uploaded by

Copyright:

Available Formats

Market Fairness: The Poor Country Cousin of Market Efficiency

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Market Fairness: The Poor Country Cousin of Market Efficiency

Uploaded by

Copyright:

Available Formats

J Bus Ethics (2018) 147:5–23

Market Fairness: The Poor Country Cousin of Market Efficiency

fairness and efficiency. In line with this dual objective,

Markets in Financial Instruments Directive 2007 What Exactly is Algorithmic Trading?

Exhibit 2 Cancel-to-trade (CTR) ratio, a proxy for algorithmic trading

Table 1 Descriptive statistics Table 1 continued

MTC (marking the close) Post-MiFID1 0.01 0.06 -0.01 0.06

Nov 2003 – Nov 2011

Spreads of 20-35 Euronext-Paris

As an explanatory variable, AT in row 7 is unrelated to

Intercept -19.467** 14.397 -0.506 -10.853 82.546 -1.251

You might also like