Alternative Trading Systems and Liquidity

Abstract

this paper is to review the importance of ATS and their impact on the liquidity at traditional market places

Full text

Alternative Trading Systems and Liquidity
by
Hans Degryse*
and
Mark Van Achter**
Paper prepared for SUERF-conference
Commission 2: Technology and Financial Markets
* Center for Economic Studies K.U. Leuven, Tilburg University and CEPR
** Center for Economic Studies K.U. Leuven
We appreciate helpful discussions with Jean-Paul Abraham, Jos Schmitt and participants at the 23
rd
SUERF Colloquium. We acknowledge financial assistance from FWO-Vlaanderen under contracts
G.0302.00 and G.0333.01.

2
I. Introduction
Traditional exchanges face enormous challenges. Technology, deregulation and
investor needs are driving forces reshaping the trading landscape throughout the
world. Technological progress not only allows direct access at traditional exchanges.
It also enables to erect new market places, called “Alternative Trading Systems”
(ATS). A general definition of ATS is “a trading mechanism developed independently
from the established market places and designed to match buyers and sellers on an
agency basis” (Salomon Smith Barney, 2001). The purpose of this paper is to review
the importance of ATS and their impact on the liquidity at traditional market places.
ATS have gained success in the U.S. after the introduction of new Order Handling
Rules in 1996 followed by “Regulation ATS” in 1998. The latter regulatory measure
mainly improves the linkages between traditional markets and ATS by requiring ATS
to become self-regulatory organization members. We will show that in the United
States ATS have been particularly successful in attracting trade in the Nasdaq dealer
market whereas they are less successful in competition with the NYSE. In Europe,
traditional trading market places automated earlier than in the United States.
Moreover, continental European exchanges are typically organized as auction systems
implying an agency nature of trading. The liquidity externality then makes it more
difficult for ATS to develop a successful business model in Europe.
The market microstructure of ATS and traditional market places is a major
determinant of their future success. This literature is mainly concerned with the
process by which investors’ latent demands are ultimately translated into transactions.
Given the different driving forces transforming the trading landscape, market
microstructure helps in judging the relative merits of the different designs of the ATS.
It also helps in making projections on their impact on liquidity at the traditional
market places.

3
The remainder of this paper is organized as follows. Section II starts by offering
evidence on the impact of technology/automation on trading costs. Section III
develops a typology of traditional exchanges and ATS. The fourth section deals with
the relation between ATS and liquidity. The final section concludes.
II. Technology and trading costs
Domowitz (2001) argues that the automation of traditional financial markets plays an
important role in the evolution of the industrial organization of the trading services
industry. He propounds that markets are firms with network externalities related to
liquidity. Although the link between liquidity and trading costs is well known, he is
the first to investigate the connection between the automation of market structure and
trading costs.
Table 1 provides an overview of the trading costs for the period 1996-1999. The
execution costs are based on data gathered by Elkins/McSherry and were published in
Institutional Investor.
1
<insert Table 1 about here>
Transactions costs are falling worldwide, illustrated by the decline of the average total
trading cost from 73 basis points in 1996 to 61 basis points in 1999. Explanations
include competition for order flow, shifts of trading strategies to accommodate
liquidity differences, more institutional trading, and pressure from new trading
systems and regulatory authorities.
Domowitz investigates whether the adoption of an automated trading technology on
traditional exchanges actually contributes to trading cost reductions. He sheds light on
1
Elkins/McSherry receive trade data on all global trades by institutional traders and compute measures
of trading costs. The data consist of average total trading costs − execution commissions, fees and
market impact (difference between the price of a stock trade and the average of that stock’s high,

4
this issue at an international level (42 countries). In this section we first briefly
summarize the methodology used in this study and its main results. Next, our own
empirical results specifically for Europe will be presented.
II.1. International empirical evidence
Domowitz uses the Elkins/McSherry data on trading costs. This allows him to make a
distinction between explicit and implicit costs. These are respectively related to
development and operating costs
2
(i.e. fees and commissions), and to the
dissemination of information on liquidity
3
(i.e. indirect trading costs like price impact
costs, including the bid-ask spread). Evidently, the various cost components may be
linked to each other. For example, minimizing price impact may imply incurring
higher commissions. The link between automation and savings in explicit trading
costs is quite obvious. Implicit trading costs, however, are at first sight not directly
related to the automation of the market structure as liquidity is only created by the
traders’ presence on the system. But obviously, an automated market system
indirectly affects liquidity as its design affects traders’ incentives and capacities to
monitor the market. Therefore, automation may shape the properties of transactions
prices and market efficiency.
Next, Domowitz tests whether total trading costs and its components depend on the
adoption of an automated trading technology by using regression-based techniques.
As control variables, volatility, turnover and market capitalization are used. The
regressions are performed on a quarterly, cross-sectional basis for all the countries in
the dataset in the period from 1996:4 to 1998:3. Some of these countries do have
exchange facilities that are largely automated with respect to execution while others
do not.
low, opening and closing prices during the day) − as a percentage of trade value for active managers
in a universe of 42 countries.
2
Trading floor development costs for instance were calculated to be two to forty times as expensive as
those for electronic market places.
3
On automated markets the electronic order books are open for insight to all clients allowing for an
optimal active liquidity management to control implicit transactions costs.

5
The results for the international sample indicate that markets that are largely
automated have average total trading costs that are, ceteris paribus, 33 to 46 basis
points lower than those of their non-automated counterparts. Both types of cost
components, explicit and implicit costs, hinge on automation. Explicit costs are
between 23 to 32 basis points lower, whereas implicit costs are 10 to 18 basis points
lower. Thus, on an international level the automated trading market microstructure
does seem to have an effect on costs. This difference might be related to the higher
floor development costs on non-automated exchanges
4
.
Domowitz (2001) discerns why the automation of markets permits the realization of
implicit cost savings. By which means do electronic market systems allow traders to
reduce price impact costs? One answer to this question is the presence of an
electronic limit order book. Via this tool that characterizes automated markets, traders
can easily and instantly monitor certain liquidity characteristics (i.e. strategic liquidity
management (see footnote 3)). This allows traders to execute their transactions when
the market is rather liquid implying a reduction of transactions costs. Indeed, in reality
the data indicate that traders do tend to use the electronic system to monitor liquidity
and trade in a strategic way using this information. The price impact of realized trades
is much smaller than that of trades executed under a naïve trading strategy that
ignores monitoring of the book and stays almost constant along different trade sizes.
II.2. Empirical evidence for Europe
The analysis of automation in Europe on the basis of the Elkins/McSherry data is
rendered somewhat trickier. The reason is that traditional European stock markets are
technologically advanced and many exchanges were already electronic since 1996,
which is the starting date of the Elkins/McSherry dataset. However, as can be seen
from Table 2, there are some notable exceptions.
<insert Table 2 about here>
4
Other reasons include industry and regulatory related matters.

6
In order to compare our results with those of Domowitz (2001), we replicate his
regression for the European countries only. More specific, we estimate the following
equation:
itititit
VolatilitytalizationMarketCapiDummyAutomationtTradingCos
β
β
β
β
3210
+++=
ε
β
β
it
it
sYearDummieTurnover +++
54
The results are displayed in table 3:
5
Table 3: Impact of electronic trading systems on trading costs in Europe
Dependent variable Total costs Implicit costs Explicit costs
Electronic trading
dummy
-7.82
(2.72)
-3.29
(1.78)
-4.52
(1.59)
# country-years 52 52 52
Legend: OLS estimates of the three proxies of trading costs (total, implicit and explicit costs) on an
“electronic trading dummy” and other control variables (market capitalization, volatility, turnover and
year dummies). Standard errors are in parentheses.
The results in table 3 show a significant negative coefficient for the electronic trading
dummy. The interpretation is that in automated markets total trading costs are about 8
basis points lower.
6
The savings in explicit trading costs are somewhat higher than
those in implicit costs. Although the results should be carefully interpreted due to the
low number of countries and the short time period, they confirm the negative
coefficients obtained in Domowitz (2001) for 42 countries. However, the magnitude
of our coefficients is substantially smaller than the conditional savings in international
trading costs due to automation as reported in Domowitz (2001). In particular, he
reported total costs savings of 33 to 46 basis points, explicit cost savings of 23 to 32
basis points, and implicit cost savings of 10 to 18 basis points. These are about 5
5
Data are taken from the FIBV website. Volatility is measured as the standard deviation of the monthly
returns on the countries’ stock index. Turnover is proxied by total trading volume divided by total
market capitalization.
6
Other specifications show that the magnitude of the coefficient is fairly stable. However, the
electronic market dummy not always turns out to be statistically significant. Moreover, the control
variables do not always show the expected sign.

7
times larger than the results obtained for Europe only, which suggests that the impact
of automation is less pronounced for this continent. Two explanations may drive these
differences. First, it is possible that the automation dummy may only capture another
step towards a full electronic market. A second explanation is that the “automation
dummy” in Domowitz may also capture agency trading or deregulatory effects.
Agency trading is dominating in (Continental) Europe even when automation was not
yet in place.
III. Typology of traditional markets and alternative trading systems (ATS)
The market microstructure literature typically distinguishes dealer markets and
auction markets. Market makers are the only providers of liquidity in dealer markets.
They are a counterparty in all transactions and quote two prices: the bid price, at
which they are willing to buy securities and the ask price, at which they will sell. The
difference between those two prices is the market maker’s spread. This spread hinges
on the degree of asymmetric information between the dealer and informed traders,
inventory costs and the remuneration for the service of providing immediacy (see
Glosten and Milgrom (1985), Ho and Stoll (1981) and Demsetz (1968), respectively).
An example of a dealer market is Nasdaq. On auction markets, on the contrary,
investors trade directly with each other or with the intervention of a broker dealer
acting as an agency trader only. All unexecuted orders are gathered in a limitorder
book. Market orders consume liquidity. Limit orders that do not execute immediately
supply liquidity and could therefore be seen as free (short-lived) options against
which market orders can be executed. Examples of auction markets are Euronext and
the Toronto Stock Exchange. Other important characteristics are the degree of
continuity of the exchanges, the degree of price discovery and the transparency (see
Madhavan (2000) for a review). Some only operate at certain points in time during the
day whereas others are continuous.
There is a wide variety in alternative trading systems (ATS). In referring to ATS we
exclude the established market places (traditional exchanges) as well as “internal

8
netting systems” (organized by individual intermediaries). A typical aspect of ATS
concerns the fact that buyers and sellers meet on an agency basis.
7
Within the ATS, we distinguish three groups of networks for which we will present a
brief description of their typical features
8
.
A first important category is Electronic Communication Networks (ECNs). Weston
(2001) describes ECNs as “electronic trading systems that allow investors to clear
trades through an open limit order book. Rather than place orders with a specialist or
dealer, traders on ECNs may anonymously submit orders and trade with each other
directly.” A typical feature is that brokers on this communication network are acting
on an agency basis only. ECNs allow traders to submit priced trades, i.e. limit orders.
Therefore, ECNs have the potential to contribute to price discovery. Most ECNs
guarantee pre- and post-trade anonymity.
A second category of ATS are external Crossing Networks. The SEC (2000) defines
crossing networks as “systems that cross multiple orders at a single price and that do
not allow orders to be crossed or executed outside of the specified times”. Crossing
systems thus only trade at scheduled times, as opposed to the continuous trading of
exchanges or other ATS. Since traders enter unpriced buy or sell orders, crossing
systems do not contribute to price discovery. Execution risk remains at crossing-
networks since the trade is not necessarily executed. The intuition is that excess
demand or supply may result. The advantage of a crossing network is that it
minimizes market impact. Trades are typically executed at the midpoint of the bid-ask
spread in the primary market. According to SSB, crossing networks cater to
institutional investors placing larger sized orders in less liquid securities. Examples of
crossing networks for Europe include ITG’s POSIT or E-Crossnet. Other crossing
systems use an auction procedure (e.g. Arizona Stock Exchange). They are similar to
the batch auctions used at traditional exchanges as they match buyers and sellers at
the same price in maximizing the matched volume.
7
A notable exception is Jiway, which allows dealers to be dual capacity traders, i.e. also to trade on
their own account.
8
The distinction between the different types of ATS is not always clearcut. For example, electronic
communication networks often also offer SORT technology.

9
A third type of ATS applies Smart order routing technology (SORT). These are
systems developed by a variety of market participants that are used to route orders to
centralized markets based on trading criteria that seek to provide best execution for
the client. This execution can be on a traditional exchange or on an electronic
communication network. The trading criteria can be price improvement or execution
speed.
ATS are evolving quickly and their future remains quite uncertain. SSB distinguishes
several business models for ATS. Some of them move to become a destination
exchange (e.g. Tradepoint into Virt-X, Archipelago). This implies that the ATS
becomes an organized market allowing them to become a destination for listed shares.
Another business model is to become a regular broker at several exchanges, i.e. a
destination broker-dealer. This essentially happens with SORT that should be able to
provide execution at several places (e.g. Instinet, a subsidiary of Reuters Company,
has become a member at 18 exchanges). It is clear that some new market places offer
several of the types of ATS discussed. For instance, ITG is offering an ECN and a
crossing network. Moreover, some of the specific aspects of ATS have been already
incorporated into the traditional exchanges (NYSE direct+ offers a crossing network).
IV. Alternative Trading Systems and Liquidity
In this section, we will first discuss US-evidence on ATS and liquidity. Next, some
evidence for Europe will be presented.
IV.1. Empirical Evidence for the United States
IV.1.1 Importance of ATS
Table 4 provides an overview of some main characteristics of the most important
ECNs in the United States. The market shares of the ECNs are presented in figure 1.

10
<insert Table 4 about here>
Figure 1: Nasdaq Share volume of ECNs
Source: Nasdaq and websites of ECNs
Most ECNs started operating in the late nineties. Nine of them are currently still
active. Jointly, they attract about 29 per cent of total share volume on Nasdaq (second
quarter of 2001), a number that has been steadily increasing from 12 per cent in the
first quarter of 1998. According to Weston (2001), two causes can be discerned for
this growth pattern. First, the changing SEC regulations are an important determinant.
For instance, the so-called “Order Handling Rules”, introduced in 1997, increased
competition because public limit orders were since then allowed to compete directly
with Nasdaq market makers. Also market makers posting orders on ECNs were since
then obligated to make those orders available for the public as well. This forced
dealers to provide greater access to ECNs for public investors. Moreover, ECNs have
been more successful in attracting trade from Nasdaq. The intuition is that ECNs offer
an agency alternative eliminating the spread charged by dealers. The NYSE is already
an auction market (with a specialist) and enjoys an incumbency advantage due to the
liquidity externality. Secondly, also the advances in technology have played a
tremendous role. As argued in the previous section, the trading systems were less
0,00%
5,00%
10,00%
15,00%
20,00%
25,00%
30,00%
35,00%
1Q98
2Q98
3Q98
4Q98
1Q99
2Q99
3Q99
4Q99
1Q00
2Q00
3Q00
4Q00
1Q01
2Q01

11
advanced compared to many European exchanges. This allows the ECNs to attract a
significant part of the market.
An interesting feature of ECNs is that important broker/dealers have become
important shareholders of ECNs. Milbourn, Boot and Thakor (1999) offer one
explanation for this evolution.
9
Although the future of ECNs is highly uncertain,
broker dealers are willing to accept relatively low returns for the moment. The
potential vital role of ECNs in the future defines these ownerships as a strategic
option. Moreover, in the action of attracting order flow by institutional investors or
retail investors, ECNs develop strategic partnerships with online brokers and
broker/dealers. This ultimately determines the success of their business plan. All
ECNs offer at least an internal limit order book. Most of them also offer SORT; that is
they route orders to either other ECNs or dealers. Two of the ECNs have an exchange
application pending. The most successful ECN in terms of Nasdaq stocks is Instinet. It
attracts about 13 per cent of Nasdaq share volume during the last quarter of 2000.
Another important player in terms of Nasdaq volume is Island with 7 per cent. The
other ECNs have a market share in Nasdaq volume of less than 2 per cent.
Next to the nine ECNs, there are some crossing networks already operating in the U.S.
and some are announced. The Arizona stock exchange organizes since 1990 single
price auctions four times a day. Its volume however is fairly low. ITG’s POSIT was
launched in 1987 and crosses seven times a day. It is the largest crossing network in
the U.S. Two very recent crossing systems are Primex Trading and Wofex. Primex
Trading exposes certain orders for auction-style competition. Prices may be away
from the best quotes in the National Market System. Wofex essentially adds SORT to
a crossing network.
IV.1.2. Market quality / liquidity
For the United States, there are already some studies describing the behaviour of
ECNs and their impact on the market quality / liquidity on traditional exchanges.
9
Note that they apply it in a different context, namely the diversification of banking activities.

12
These include the following : Huang (2000), Hendershott and Mendelson (2000),
Simaan, Weaver and Whitcomb (1998), Domowitz (2001), Barclay, Hendershott and
McCormick (2001) Weston (2001), Conrad, Johnson, and Wahal (2001), Benhamou
and Serval (2000), Domowitz, Glen and Madhavan (2001), Domowitz and Steil
(1999) and Naes and Odegaard (2001). In most of these studies, the traditional market
under consideration is Nasdaq as ECNs have proven to perform best for shares noted
on this exchange (cfr. infra). In this subsection, we will briefly describe and compare
the main results of some of these studies. This will be done focusing on four aspects
of market quality/liquidity, namely bid-ask spreads, depth of the market,
informational efficiency and price discovery.
1) Bid-Ask Spreads
Weston (2001) investigates whether the increased market share of ECNs leads to
tighter spreads (monthly average quoted, effective and relative spreads for stock i in
month t), i.e. whether ECNs have a significant negative impact on spreads on
traditional markets. For this purpose, he performs the following regression using a
long time-series and large sample of firms
10
:
(
)
(
)
(
)
(
)
(
)
itititit
i
it
SizeormsECNshareECNdummySpread lnReflnlnln
4321
ββββ
α
++++=
(
)
(
)
(
)
ε
βββ
it
itit
it
adesnumberoftrVolatilityTurnover ++++
765
lnln
The variables “ECNdummy”, indicating ECN activity in a given stock-month, and
“ECN market share” allow to test the effect of ECN activity on spreads. The variable
“Reforms” is included to capture possible spread effects of any market reforms (i.e.
Order Handling Rules). The independent (control) variables in this model were
chosen according to Stoll (1997) and Wahal (1997). They are used to capture well-
known determinants of bid-ask spreads, and of execution costs in general. For
instance, the selected size variable controls for the fact that orders that are large
10
This is a multivariate fixed-effect model that allows for within-firm variation in the parameters to
account for unobserved heterogeneity in liquidity for the sample of firms.

13
relative to normal trading volume are likely to have higher execution costs because of
adverse selection effects. Log transformations of these variables are used to reduce the
skewness.
The
β
1
- and
β
2
-coefficients are of interest to us and are consistently negative and
statistically and economically significant for all specifications (i.e. for the three kinds
of spreads). This implies that ECNs induce competitive pressure on the Nasdaq
market. The exact number for the
β
1
-coefficient in the average quoted spread
regression is equal to -0.0041, implying a 4 percent ceteris paribus decrease for this
spread measure. For the effective and the relative spread, this decrease amounts 10
and 7 percent respectively. The
β
2
-coefficient indicates that a one percent increase in
ECN activity lowers the average quoted spread by 0.714 percent. For the effective and
the relative spread, this decrease amounts 0.917 and 0.07 percent respectively. Weston
argues these results are particularly strong because the data used actually give an
underestimation of the true impact due to the manner in which volumes are reported
to the Nasdaq (cfr. supra). Note, however, that they are only valid for small trades, not
for block trades.
Thus, in addition to regulatory market reforms, the growth of ECNs has helped to
significantly lower trading costs. As such, it has mitigated the negative effects of the
suspected imperfect competition among Nasdaq dealers (e.g. Huang and Stoll (1996),
Christie and Schultz (1994), Weston (2000))
11
.
Domowitz (2001) constructs an American sample by gathering data from institutional
investors. For this dataset, total trading costs for executions by institutional investors
through ECNs and through traditional brokers and markets are compared
12
. Calculated
yearly savings from 1993 through 1996 using automated systems vary from 31 to 65
percent, relative to trades executed by traditional brokers or dealers.
13
Domowitz even
manages to invalidate the conventional wisdom that automated trading venues are
11
This is supposed to be due to practices such as payment for order flow and preferenced trading used
by traditional dealers to attract order flow through non-price competition. Thus, large spreads are
prevented from from being competed away (Weston (2001)).
12
Note that total trading costs also include price impact, determined as a geometric average of realized
and effective spreads, and measured relative to short-run industry performance.
13
Average savings amount to 46 percent.

14
cheaper only because “easier” trades are more often sent to them as he proves that
even for more difficult trades, savings from automated execution are evident
14
.
This empirical evidence is also consistent with Conrad, Johnson and Wahal (2001),
although they use a somewhat different approach. They determine what the difference
in realized execution costs is between external crossing systems (POSIT or an after-
hours cross on Instinet), ECNs (Instinet) and traditional markets (NYSE, Amex or
Nasdaq). These three trading systems are engaged in a competition for order flow. In
their dataset
15
, the distinction is made between single and multiple mechanism orders,
which are respectively orders that are completely executed by a single trading system
(91 percent of all orders) and those in which trades are filled by more than one trading
system (9 percent of all orders). Note that there is considerable time series variation,
but no trend in the distribution of single mechanism orders. Further, the data show
substantial differences in size between orders executed on the three mechanisms.
Order fill rates are lowest for crossing systems as it concerns a mere function of
liquidity on the system (cfr. contra-side depth), which is exogenous to the trader. As
traders on ECNs and on traditional broker systems can trade anonymously, they
endogenously increase the probability of a fill. Evidently, multiple mechanism orders
have the largest execution costs, as they are most difficult to fill.
As in Domowitz (2001), total execution costs are measured as the sum of implicit and
explicit costs. Obviously, comparing execution costs between different trading
systems univariately can be quite misleading as the trading mechanisms may represent
varying degrees of aggressiveness on the part of the institution
16
. One needs to take
the differences in order characteristics between these systems into account. For
instance variation in order difficulty and other characteristics influencing liquidity and
thus trading costs. These are controlled for using two methods, i.e. a “matched-
14
Domowitz defines more difficult trades as having above median values of trade size and volatility, or
having below average market capitalization (firm size), i.e. the controls used above.
15
Note that only to describe ECN activity, only data for Instinet were used as the remaining ECNs only
commenced operations after the end of their sample period.
16
Conrad et al. (2001) offer the following ranking on aggressiveness : external crosses < ECN-
executions < broker-dealer operations. These differences result in a natural sorting of order difficulty
across the categories.

15
sample” approach
17
and a regression-based approach
18
as in Weston
19
. Both these
methods yield quite similar results. Compared to traditional brokers, execution costs
on crossing systems are substantially lower. For ECNs, this cost advantage is even
more pronounced. Note that these results are quite robust and that the differences can
be primarily attributed to distinct implicit costs.
Conrad, Johnson and Wahal (2001) note, however, that an endogeneity problem may
arise as the choice of trading mechanism could be endogenous to (ex post) realized
execution costs. Orders that are more difficult to fill, and thus incur higher ex post
execution costs, are more likely to be sent to mechanisms guaranteeing a high fill rate.
This issue, which leads to inconsistent estimates, is not accounted for in the above
mentioned methods and therefore needs to be addressed by using a two-stage
procedure (“endogenous switching regression method”) following Madhavan and
Cheng (1997). The cost differentials described above seem to persist when applying
this model, in fact they are even more pronounced.
2) Depth of the Market
Besides performing a bid-ask spread comparison, Weston (2001) also investigates
whether the increase in ECN market share leads to greater depths. For this purpose, he
performs the following regression:
(
)
(
)
(
)
it
a
it
a
it
a
it
aa
it
volatilitypricevolumeyECNactivitDepth lnlnln
43210
++++=
(
)
itt
a
it
a TimeDummyentrationMarketConc
ε
+++
65
ln
17
Which controls for trade direction, order instruction, order size, exchange listing and market
capitalization without imposing any functional form restrictions.
18
Control variables : order size, inverse of stock price, logarithm of market capitalization, exchange
listing, return volatility, cumulative size-decile adjusted return, institution-specific indicator
variables, indicator variables for external crosses and ECN-executed orders.
19
Note that another possibility for comparing execution costs is focusing on multiple mechanism
orders, as order characteristics by definition are held constant across the trades. Also the investor
chooses how to break up the order, and where and in what sequence to place the order.

16
The presence of an ECN does seem to increase the quoted depth ceteris paribus by
11,6 percent. A one percent increase in ECN activity improves quoted depth by 0,27
percent all other variables held constant. So ECN activity improves the total quality of
the market. These conclusions, however, are disputed by Barclay, Hendershott and
McCormick (2001) who study transactions data for June 2000 and conclude that ECN
trading lowers quoted depths.
3) Informational Efficiency
Weston (2001) suggests that ECNs do impose higher adverse selection costs on
traditional markets through more anonymous trading
20
. An increase in anonymity
through ECN trading may therefore increase information costs, urging intermediaries
to charge larger spreads (Amihud and Mendelson (1986), Glosten and Harris (1988)).
So, although ECNs lower trading costs (cfr. supra), they reduce the informational
efficiency of prices. Note that this conjecture does not hold if the ECN functions as a
separate market. In this case the presence of an ECN reduces the amount of
information asymmetry in a dealer market by providing an alternative venue for
information-based trades. Weston performs a test on the change in anonymity of
trading on the Nasdaq due to ECN trading, i.e. estimating the adverse selection
component of spread (Huang and Stoll (1997)) and regressing this measure on the
level of ECN activity and a group of control variables
21
. An increase in adverse
selection costs linked to ECN trading is noticed, confirming the first conjecture stated
above. However, these costs are outweighed by benefit of lower overall transaction
costs.
4) Price Discovery
Conrad, Johnson and Wahal (2001) describe the link between the efficiency of the
primary markets’ price discovery mechanism and the success of ECNs. For the United
20
Intermediaries face uncertainty on the type of trader they deal with, i.e. informed or uninformed
ones.
21
These control variables include market capitalization, share turnover, return volatility and market
concentration, and are also suspected to affect information costs.

17
States, it has been extensively proven that transaction costs are significantly lower on
the NYSE than on Nasdaq (for example Hasbrouck (1995), Huang and Stoll (1996)).
An obvious rationale for this difference is the distinction in trading mechanisms that
are employed on both markets, i.e. auction markets provide more adequate price
discovery than the dealership markets. In their study, they refer to Hendershott and
Mendelson (2000), who state two necessary conditions for crossing systems to be
successful when co-existing with a dealer market. Firstly, as these systems do not
provide active price discovery themselves, they need to rely on a primary market
providing an adequate price discovery mechanism. Secondly, the crossing network
initially needs to attract at least a minimum threshold of volume from this primary
market so that the pool of liquidity is sufficiently large
22
. Based on these conditions,
one could postulate that crossing networks will be more successful in competing for
NYSE shares and therefore primarily focus on listed securities. ECNs on the other
hand, engage themselves in active price discovery, and will therefore rather compete
with primary markets with higher transaction costs and fragmented order flow
23
. In
fact, their success is inversely related to the efficiency of the primary market, i.e. if
bid-ask spreads are higher on the primary market, ECNs become a truly competitive
alternative
24
. Clearly, external crossing systems and ECNs compete for order flow in
different dimensions as certain clientele effects arise. Empirical evidence seems to
support both these conjectures as 90 percent of all orders executed on external
crossing systems are for NYSE securities and 80 percent of all ECN-executed orders
are for Nasdaq securities (sample by Conrad, Johnson and Wahal (2001)).
IV.2. Empirical Evidence for Europe
22
Referring to the Hendershot and Mendelson paper, Conrad et al. quote that “Volume on crossing
systems that provide no price discovery function has a natural upper bound since the system cannot
exist independent of the primary price-setting mechanism, whether it be an auction or dealer market.
To the extent that other systems (such as ECNs) provide a price discovery mechanism, they can exist
and grow independently.”
23
ECNs do make a significant contribution to price discovery and therefore do not necessarily engage
free-riding off of price discovery by traditional dealers on Nasdaq (Huang (2000)).
24
Note that a major determinant of the higher bid-ask spread on Nasdaq is the difference in anonymity,
i.e. the Nasdaq market structure is more anonymous than the NYSE (Garfinkle and Nimalendran
(1998) en Heidle and Huang (2000)) leading to higher adverse selection costs and thus to higher
spreads.

18
It is a much more difficult exercise to gauge the importance of ATS in comparison to
the European exchanges. ECNs, like Instinet, are brokers/dealers allowing investors
to trade on several European exchanges. The two most prominent ECNs that are
active on the European market are Tradepoint and Jiway.
1) Tradepoint
Tradepoint recently merged with Swiss Exchanges into Virt-X, which is an
attempt to create a pan-European blue chip exchange. Using
unprecedented technology, their aim is to become competitive by
providing the scope for significant reduction in cross border transaction
costs at each stage of the trading, clearing and settlement process. Trading
on a sectoral base is encouraged, rather than trading on a national base.
Their aim is to “capture ten percent of the pan-European blue chip trading
within twelve months” (cfr. site www.virt-x.com). Actually the market’s
structure (i.e. a continuous electronic public limit order book with opening,
intra-day and closing single price auctions and full anonymity and
facilities to support liquidity providers and off book and block trading
requirements) strongly resembles the one offered by auction markets (cfr.
infra).
2) Jiway
Jiway, an initiative of OM Grüppen and Morgan Stanley Dean Witter
combines a limit order book and market makers
25
. Its major focus is giving
retail investors greater access to European and American stock markets.
Thus it aims at small orders and tries to internationalize the retail market
so as to improve liquidity on these markets.
Next to the ECNs, there are at least two crossing networks active in European stocks,
i.e. ITG-Europe’s POSIT and E-Crossnet. It is difficult to obtain estimates of their
25
Note that meanwhile (as of September 2001), M.S.D.W. sold their stake integrally to OM Grüppen.

19
activity. However, in sum and up to now, ATS in Europe are far less important than in
the United States.
1) ITG-Europe’s POSIT
Since 1998, institutional investors and broker dealers can trade on ITG
Europe’s POSIT, a crossing system that is active in shares of eight European
countries. Anonymity is guaranteed to reduce market impact.
2) E-Crossnet
Since 1999, E-Crossnet operates in 14 European countries and also aims at
institutional investors and broker dealers. Its objectives, structure and dealing
mechanism are roughly comparable to those of POSIT.
The empirical evidence on the interaction between ATS and market quality for
Europe is rather scarce. Board and Wells (1999) offer a comparison of SETS
(traditional exchange) and Tradepoint (ECN) concerning liquidity and best execution
of UK shares. Therefore they compare prices available on those two exchanges, in
fact the extent of price improvement opportunities is measured and analyzed.
26
Their
analysis indicated that while SETS was clearly more active during the period under
consideration, Tradepoint managed to offer better prices for between 45 and 90
minutes per trading day, at volumes that were roughly comparable to those offered by
SETS. The reason why they still did not manage to attract sufficient trading volume,
although being cheaper, is attributed to insufficient depth. Board and Wells propound
that “if the other ECNs that are operating or planning to operate display similar results
as Tradepoint, and particularly if they attract significant business, then there will be
significant periods of the day in which the SETS price is not the most attractive
price.”
26
The comparison is executed towards some specific factors, e.g. the availability of best prices on the
two markets, the spread on each market, available depth at best prices, etc.

20
Note that most of the empirical evidence for the impact of ECNs on traditional
exchanges for the United States concerns Nasdaq (cfr. supra), which is a dealer
market. Therefore, the stated findings and insights do not necessarily apply for Europe
where most of the stock markets offer an electronic auction-based mechanism. As this
sort of trading system is quite analogous to the one offered by ECNs, they will
probably even face difficulties in attracting trade volumes. Clearly, the traditional
markets, acting as incumbents, enjoy a major liquidity externality, implying
difficulties for ECNs in capturing a market segment of their own. Moreover, it is clear
that ECNs have not flourished to such an extent in the European markets so far,
because the traditional exchanges have been proactive in addressing the changing
needs of investors, i.e. in creating efficient trading facilities themselves.
Intuitively, we expect crossing networks to be relatively more successful than ECNs
for Europe. To reach this conclusion, we extend the Conrad, Johnson and Wahal
(2001) results on price discovery to the European “auction market” case.
Unfortunately there is no evidence reported so far concerning the magnitude of
trading on crossing networks. Initiatives as E-Crossnet and ITG Europe’s POSIT (cfr.
supra), however, demonstrate that these kinds of networks can indeed be erected.
Given these arguments, one could expect the traditional markets to create their own
passive call market in the future, parallel to their own market.
In the market microstructure literature, previous studies have investigated whether the
trading activity of a dually-traded stock on one market has an effect on trading
activity on the other one (and thus not necessarily on the spread). Pagano and Röell
(1991) initiated this research methodology and investigated whether trading of Italian
equities on SEAQ International implied trade diversion or trade creation for the Milan
Stock Exchange. This methodology has been replicated by e.g. Anderson and Tychon
(1993) and Degryse (1996) for the impact of SEAQ International on Belgian equities.
We will now apply this methodology to test whether a variable related to trading
activity on an ECN (Virt-X) helps to explain trading volume on the “local” exchange
(Paris Bourse). Trading on ECNs may have displaced activity from the local exchange
to the ECN. Alternatively, it may also have generated a stimulus in trading as some
institutional investors find the source of or outlet for the shares dealt at the ECN. Our

21
dataset consists of weekly trading data for an eight-month period on the Paris Bourse
and on Virt-X. Functioning as dependent variable for our regressions is the volume on
the Paris Bourse exchange for ten randomly selected dually traded stocks that are all
members of the CAC40 index
27
. Next to the Virt-X volume variable, explanatory
variables are lagged values of the dependent variable, current and lagged values of
total market volume, current and lagged values of the average return and the volatility
of the relevant stock (respectively measured by the monthly average and the standard
deviation of daily returns)
28
. These are included to control for other possible
determinants of trading volume, a choice that is based on Pagano and Röell (1991). A
negative and significant coefficient for the Virt-X volume variable is interpreted as a
symptom of trade diversion from the Paris Bourse to Virt-X. A positive significant
one indicates trade creation. Most of our regressions (9 out of 10), however, generate
a negative but insignificant coefficient indicating no effect at all. Note that this is
probably due to the small time span of the sample. We expect the effect to increase,
within certain boundaries (cfr. supra), as Virt-X will continue being operative. Clearly
a period of almost three months, in which its introduction to the financial markets
occurred, is rather short to state any strong conclusions on its impact. Moreover, our
estimated coefficients could also capture some sort of “summer effect”, as the
introduction of Virt-X coincides with the summer break. This obviously limits our
results.
V. Concluding Remarks
The purpose of this paper was to discuss the relationship between alternative trading
systems (ATS) and liquidity. Two important trends can be distinguished. First, ATS
are currently more successful in the United States than in Europe. Second, within the
United States, there exists an interesting divergence between the impact of ATS on
the Nasdaq dealer market and on the NYSE. ATS are attracting about 30 per cent of
27
Namely for Alcatel, Aventis, Axa, Carrefour, Eurotunnel, France Telecom, Orange, Renault, Usinor
and Vivendi.
28
Note that replacing the Virt-X volume variable by a dummy variable as in Pagano and Röell (1991)
does not change our results.

22
market share in the Nasdaq market, whereas their impact on the NYSE is rather small.
Trading volume on ATS in Europe is currently still marginal compared to the
established market places.
Two forces explaining these differences should be distinguished. The first is that
European traditional market places were earlier automated than their American
counterparts. International evidence shows that automation reduces transaction costs
considerably. ATS are the exponent of automated systems and should therefore be
more successful in the United States. Our empirical work shows that automation also
has a significant impact on trading costs in Europe, but still less substantial than in an
international context. This observation brings us to a second explanation, i.e. the
agency nature of trading. European markets are mostly organized as an auction market
where traders can submit market and limit orders. ECNs allow investors to trade with
each other via a limitorder book without the intervention of a dealer. This market
microstructure is close to the one of incumbent European exchanges (e.g. Euronext).
Therefore ECNs are successful in attracting Nasdaq trading volume and are expected
to be less successful in competition with the NYSE or European exchanges. Crossing
networks are more successful in realizing trades of NYSE listed securities. This leads
us to the projection that crossing networks may be a more successful ATS business
model in Europe than ECNs.
Several studies discuss the impact of ATS on the market quality/liquidity of American
markets. Bid-ask spreads seem to decrease due to competition of electronic
communication networks. Thus competition seems to be more important than
fragmentation of markets. The results on market depth are inconclusive. ECNs reduce
the informational efficiency of the market. The reason is that ECNs typically allow for
anonymous trading, leading to an increase in the adverse selection component of the
spread. Crossing networks rely on price discovery at the primary exchange while
ECNs actively contribute to the price discovery process.
Currently, trading volume on alternative trading systems in Europe is rather low
compared to the established market places. Consistent with this, our empirical work

23
does not reveal significant trade diversion or trade creation effects of Virt-X on the
incumbent European exchanges. Evidence from the interaction between Tradepoint
and SETS shows that ATS may face a problem of market depth in Europe.

24
REFERENCES
Amihud, Y. and Mendelson, H. (1986) Asset pricing and the bid-ask spread, Journal
of Financial Economics 17, pp. 223-249.
Anderson, R. and Tychon, P. (1993) Competition among European financial markets:
the case of cross-listed Belgian equities, IRES discussion paper nr. 9314.
Barclay, M., Hendershott, T. and McCormick, T. (2001) Electronic communications
networks and market quality, working paper, University of Rochester.
Benhamou, E. and Serval, T. (2000) On the competition between ECNs, stock
markets and market makers, Financial Markets Group discussion paper.
Board, J. and Wells, S. (1999) Liquidity and best execution in the UK: a comparison
of SETS and Tradepoint, Financial Markets Group discussion paper.
Conrad, J., Johnson, K. and Wahal, S. (2001) Alternative trading systems, working
paper.
Christie, W. and Schultz, P. (1994) Why do Nasdaq market makers avoid odd-eighth
quotes?, Journal of Finance 49, pp. 1813-1840.
Degryse, H. (1996) Competition between auction and dealership markets: Brussels
versus London, Revue de la Banque/Bank-en Financiewezen, pp. 378-386.
Demsetz, H. (1968) The cost of transacting, Quarterly Journal of Economics 82, pp.
33-53.
Domowitz, I. (2001) Liquidity, transaction costs, and reintermediation in electronic
markets, working paper.
Domowitz, I., Glen, J. and Madhavan, A. (2001) Global equity trading costs, working
paper.
Domowitz, I. and Steil, B. (1999) Automation, trading costs, and the structure of the
securities trading industry, Brookings-Wharton papers on Financial Services.
Glosten, L. and Harris, L. (1988) Estimating the components of the bid/ask spread,
Journal of Financial Economics 21, pp. 123-142.
Glosten, L. and Milgrom, P. (1985) Bid, ask, and transaction prices in a specialist
market with heterogeneously informed traders, Journal of Financial
Economics 14, pp. 71-100.

25
Hasbrouck, J. (1995) One security, many markets: determining the contributions to
price discovery, Journal of Finance 50, pp. 1175-1199.
Hendershott, T. and Mendelson, H. (2000) Crossing networks and dealer markets:
competition and performance, Journal of Finance 55, pp. 2071-2116.
Huang, R. (2000) Price discovery by ECNs and Nasdaq market makers, working
paper, University of Notre Dame.
Huang, R. and Stoll, H. (1996) Dealer vs. auction markets: a paired comparison of
execution costs on Nasdaq and the NYSE, Journal of Financial Economics 41,
pp. 313-357.
Ho, T. and Stoll, H. (1981) Optimal dealer pricing under transactions and return
uncertainty, Journal of Financial Economics 9, pp. 47-73.
Madhavan, A. (2000) Market microstructure: a survey, Journal of Financial Markets
3, pp. 205-258.
Madhavan, A. and Cheng, M. (1997) In search of liquidity: block trades in the
upstairs and downstairs markets, Review of Financial Studies 10, pp. 175-203.
Milbourn, T., Boot, A. and Thakor, A. (1999) Megamergers and expanded scope:
theories of bank size and activity diversity, Journal of Banking and Finance
23, 195-214.
Naes, R. and Odegaard, B.A. (2001) Equity trading by institutional investors: to cross
or not to cross?, working paper, Norwegian School of Management.
Pagano, M. and Röell, A. (1991), Dually-traded Italian equities: London vs. Milan,
CEPR discussion paper nr. 564.
Salomon Smith Barney (2001) Trading ip: The new world of electronic trading,
research report.
Securities and Exchange Commission Report (S.E.C.) (2000).
Simaan, Y., Weaver, D. and Whitcomb, D. (1998) The quotation behavior of ECNs
and Nasdaq market makers, working paper, Baruch College.
Stoll, H. (1992) Principles of trading market structure, Journal of Financial Services
Research 6, pp. 75-106.
Wahal, S. (1997) Entry, exit, market makers, and the bid-ask spread, Review of
Financial Studies 10, pp. 871-901.
Weston, J. (2000) Electronic communication networks and liquidity on the Nasdaq,

26
working paper.
Weston, J. (2001) Competition on the Nasdaq and the growth of electronic
communication networks, working paper.

27
Table 1: Total trading costs in 42 countries
Country 1996
(basis points)
1997
(basis points)
1998
(basis points)
1999
(basis points)
Argentina 93.1 59.7 48.7 62.7
Australia 57.0 53.8 47.0 54.2
Austria 40.3 39.9 54.1 42.7
Belgium 37.1 31.1 33.9 27.9
Brazil 63.1 53.7 46.6 47.1
Canada 63.0 51.6 43.9 39.9
Chile 115.4 60.4 47.0 131.0
Colombia 99.6 91.7 95.1 96.3
Czech republic 134.6 150.5 161.0 71.2
Denmark 35.7 45.4 43.4 41.1
Finland 41.9 42.3 44.0 40.7
France 29.9 26.7 26.6 24.9
Germany 39.3 33.3 27.6 28.7
Greece 64.4 66.9 63.6 87.3
Hong kong 59.2 56.6 50.1 43.7
Hungary 145.2 163.7 102.3 71.6
India 85.8 65.0 64.8 128.7
Indonesia 108.5 92.4 95.5 84.8
Ireland 153.3 105.1 99.4 71.9
Italy 36.1 29.7 30.4 34.2
Japan buy 30.6 26.5 18.2 25.1
Japan sell 56.0 47.1 36.3 25.1
Luxembourg 75.5 73.0 70.0 102.3
Malaysia 87.3 87.8 90.8 90.7
Mexico 69.3 54.7 61.0 55.6
Netherlands 69.3 25.8 30.0 28.4
New Zealand 53.6 38.5 38.9 35.3
Norway 46.1 34.0 36.4 34.4
Peru 93.9 80.1 76.0 89.6
Philippines 114.9 107.5 105.0 109.0
Portugal 62.7 59.9 41.1 42.7
Singapore 71.9 76.6 84.9 64.9
South Africa 89.6 68.3 58.5 80.1
South korea 228.9 200.1 97.8 78.9
Spain 47.1 34.9 43.0 42.3
Sweden 36.1 30.6 30.9 31.5
Switzerland 37.1 44.0 46.0 36.5
Taiwan 72.9 66.5 56.8 54.0
Thailand 93.8 87.2 75.5 82.6
Turkey 77.2 68.4 57.1 40.5
Uk buy 73.7 75.1 71.0 71.1
Uk sell 32.8 30.1 34.2 30.5
Us nyse 34.1 31.5 23.6 24.6
Us otc 51.9 39.0 29.9 33.3
Venezuela 113.4 158.4 144.7 195.8
Average
73.2 65.9 59.6 60.8
Source: Institutional Investor

28
Table 2: Electronic Trading Systems in Europe
European Exchange Electronic since
Amsterdam 1994
Austria 1999
Borsa Italiana 1994
Brussels 1996
Copenhagen 1999
Deutsche Borse 1992
Finland 1997
London Stock Exchange 1997
Madrid 1989
Oslo 1999
Paris Bourse 1988/1994
Stockholm 1989
Switzerland 1996
Source: Internet and Salomon Smith Barney

29
Table 4: ECN Characteristics (US)
ECN Archipelago ATTAIN Bloomberg
Tradebook
BRUT/Strike Instinet Island MarketXT NEXTRADE REDIBOOK
Starting date 01/97 02/98 12/96 05/98 1969 01/97 01/00 11/98 11/97
Ownership
by strategic
partners
Yes Not yet No Yes Reuters Yes - - Yes
Strategic
partnerships
Tradepoint - - - Yes - Yes Yes -
Technology Internal book
SORT
Plans to form
exchange
Internal book Internal book
SORT
Agency broker
Internal book
SORT
Internal book
Agency broker
Block trades
Internal book Mainly after
hours trade
SORT
Internal book
SORT
Exchange
application
pending
Internal book
SORT
Trade volume
as % of
Nasdaq (last
quarter 2000)
1.9% 0% 1.3% 1.8% 13% 7.1% 0% 0% 1.7%
Source: internet