Short- and Long-Term Effects of Multimarket Trading
We analyze short- and long-term effects of multimarket trading by examining the entries of multiple markets into transacting three ETFs, DIA, QQQ, and SPY. We find that large-scale entries improve overall market quality, while small-scale entries have ambiguous effects. Our results show that the competition effect dominates the fragmentation effect over a long horizon and that market fragmentation leads to a decline in trading costs. Further, we find that the order handling rules help mitigate the fragmentation effect and facilitate the competition effect. We do not find that multimarket trading harms price efficiency or increases price volatility. Copyright 2007, The Eastern Finance Association.
Electronic copy of this paper is available at: http://ssrn.com/abstract=958176
Short- and Long-Term Effects of Multimarket Trading
Fairleigh Dickinson University
Bonnie F. Van Ness
University of Mississippi
Robert A. Van Ness
University of Mississippi
Current version: January 5
Robert A. Van Ness
University of Mississippi
School of Business
329 Holman Hall
University, MS 38677
The authors would like to thank two anonymous referees, Arnold Cowan, Ken Cyree,
John Conlon, Wenbin Tang, and Bidisha Chakrabarty and participants at the 2006
Financial Management Association Annual Meeting for their valuable comments and
suggestions. All errors are our own responsibility.
Electronic copy of this paper is available at: http://ssrn.com/abstract=958176
Short- and Long-Term Effects of Multimarket Trading
We analyze both short- and long-term effects of multimarket trading by examining the
entries of multiple markets in three actively traded Exchange Traded Funds (DIA, QQQ,
and SPY). Using a time series analysis, we follow the market evolution of these ETFs
with regard to order flow fragmentation, trading activity, trading costs, volatility, price
efficiency, and price impact. We find that large scale entries, mainly the NYSE, improve
overall market quality while small scale entries have ambiguous effects. We find
evidence that the competition effect dominates the fragmentation effect over a long-term
horizon. Market fragmentation is associated with declines in trading costs. We also
present evidence against the view that multimarket trading harms price efficiency and
increases price volatility. Additionally, we find that the order handling rules help mitigate
the fragmentation effect and facilitate the competition effect.
JEL Classification: G14; G18
Key Words: Exchange Traded Funds, Multimarket Trading, Competition, Fragmentation,
Multimarket trading is prevalent in U.S. equity markets as exchange listed
securities can be traded on various markets. There are numerous studies on the effects of
multimarket trading. Yet, the results are controversial. Previous studies are centered on
two opposite effects of multimarket trading. On one hand, multimarket trading enhances
competition, improves liquidity, fosters innovation, and caters to traders’ diverse needs
(e.g. Harris, 1993 and Macey and O'Hara, 1999). On the other hand, multimarket trading
induces fragmentation, which leads to a decrease in order exposure, difficulty in
maintaining price/time priority, and impedes the price discovery process (Mendelson,
1987). To date, studies on the competition and fragmentation effects of multimarket
trading focus only on the short-term effects (e.g., Battalio, 1997; Bennett and Wei, 2003;
Boehmer and Boehmer, 2003). Using trade and quote data from the initial listing of the
DIA, QQQ, and SPY, we follow the evolution of the market for these exchange traded
funds (ETFs). The new market entrants may not overcome the liquidity externalities in
the short-term and hence fragmentation and competition effects may not be clear. As the
entrants gradually increase their market share, these effects are more visible. Hence, a
long-term study of the effects of multimarket trading is warranted.
The DIA, QQQ, and SPY are three heavily traded ETFs, which began trading
primarily on the AMEX and NASDAQ and were sequentially joined in trading by the
NYSE, regional stock exchanges, and Electronic Communications Networks (ECNs).
These ETFs are traded in up to nine markets during our study time period.
We acknowledge that our findings on the three ETF’s may not extend to other securities. ETFs are
diversified securities (as opposed to a firm’s common stock), and hence our findings regarding these three
ETFs may differ from findings regarding common stock. The ETFs track an index that constitutes a basket
of securities. Subrahmanyam (1991) and Gorton and Pennacchi (1993) suggest that basket securities have
market for these ETFs evolves, so does the nature of competition for order flow and the
level of fragmentation. We use a time series analysis to study the effects of market
evolution on order flow fragmentation, trading activity, trading costs, volatility, price
efficiency, and price impact.
The results of our study show that the new markets are able to overcome the
liquidity externalities enjoyed by the primary market. As a result, the market for ETFs is
highly dispersed. The short-term effects of a market entry depend on the size of the new
market. Large scale entries, mainly the NYSE, improve the overall market quality while
small scale entries have ambiguous effects. A long-term effect of multimarket trading is
a decline in trading costs. We find that fragmentation and price volatility are not related
in the multivariate analysis. In addition, there is no evidence that multimarket trading
impairs price efficiency. We suggest that the Order Handling Rules (OHRs), an attempt
by the SEC to make the market more transparent, may have mitigated the fragmentation
effect and enhanced the competition effect of multimarket trading.
As previous studies examine only a particular market entrance and a short time-
span, such studies do not provide adequate information to generalize the effects of
multimarket trading. We believe a long-term study of multimarket trading is warranted
to assist the trading venues and regulatory bodies in determining the most efficient
market structure. On April 6, 2005, the Securities and Exchange Commission (SEC)
approved Regulation National Market System (NMS), which includes applying price-
priority across all markets. This rule targets the creation of a virtual price-priority
Central Limit Order Book to link all markets. Opponents of the new rule state that the
lower adverse selection costs and low cost diversification possibilities. Since basket securities provide
diversification and have lower adverse selection costs for portfolio trading, the market for ETFs attracts a
high proportion of liquidity traders.
extent of market fragmentation and its adverse effects are probably overstated (Stoll,
The paper proceeds as follows. In section 2, we provide a brief summary of
related studies and develop hypotheses. Section 3 describes the data and the
methodologies used in this paper. Section 4 contains empirical analyses and results.
Section 5 concludes.
2. Related studies and hypotheses
Exchanges may use Unlisted Trading Privileges (UTP), which allows them to
trade securities listed on other exchanges, or they may trade dually-listed companies that
have their primary listings on other exchanges. With dual listings and UTPs, securities
can trade on multiple exchanges. Trading on multiple exchanges is referred to as
Research that analyzes liquidity-based competition for order flow in a
multimarket setting offers contradictory predictions about the effects of multimarket
trading. Mendelson (1987), Hendershott and Mendelson (2000), and Parlour and Seppi
(2003) suggest that there is a tradeoff between fragmentation and consolidation.
Mendelson (1987) argues that fragmentation, on one hand, increases the price variance
faced by individual investors. On the other hand, fragmentation reduces trading costs
because of inter-dealer competition. Hendershott and Mendelson (2000) find that the
competition of a passive crossing network (CN) with an existing traditional dealer market
can either increase or decrease welfare, depending on the security traded and trader
Further, Parlour, and Seppi (2003) show that competition between
exchanges, either as new markets open or as firms cross-list stocks, can increase or
decrease aggregate liquidity relative to a single market.
In contrast to these studies, Pagano (1989), Chowdhry and Nanda (1991),
Bernhardt and Hughson (1997), and Madhavan (1995) suggest that multimarket trading
induces a negative effect. Pagano (1989) suggests that fragmentation is Pareto-inferior to
consolidation. As a market becomes fragmented the positive network externalities of
consolidation are diminished. Chowdhry and Nanda (1991) show that fragmentation
leads to reductions in liquidity because fragmented markets allow informed traders to
exploit their private information more effectively by splitting trades across multiple
venues. All else equal, the model of Chowdhry and Nanda shows that fragmentation
reduces trading volume on the central market. Bernhardt and Hughson (1997) document
the possibility that splitting orders between market makers causes the market makers to
set less competitive prices, thereby increasing trading costs. Madhavan (1995) suggests
that fragmentation leads to higher price volatility and violations of price efficiency.
Empirical evidence regarding the impact of multimarket trading is also
inconsistent. Battalio (1997) finds that bid-ask spreads narrow when a third market
broker-dealer enters the market for NYSE-listed securities. Wahal (1997) documents a
negative relation between bid-ask spreads and the number of dealers, indicating that
competition enhances market quality. De Fontnouvelle, Fishe, and Harris (2003)
examine competition among option exchanges and find that competition leads to lower
The CN’s low cost attracts new liquidity traders who would not otherwise trade and the increase of order
flow from liquidity traders helps reduce adverse selection costs and leads to narrower spreads. However,
some traders may use the dealer market as the last resort, thereby increasing the riskiness for the dealer
market and causing dealers to set wider spreads. Hence, the overall effect is unclear.
trading costs. Boehmer and Boehmer (2003) analyze the trading of SPY, DIA, and QQQ
and find that all measures of trading costs decline while quoted depth increases following
the entrance of the NYSE in the market. In contrast, Arnold, Hersch, Mulherin, and
Netter (1999) find that consolidating trades into a single venue leads to lower spreads.
Bennett and Wei (2003) provide evidence that trading costs and price volatility decline,
and price efficiency improves when firms move from the NASDAQ to the NYSE. They
attribute the overall improvement in market quality to the more consolidated system of
the NYSE. Hendershott and Jones (2005) look at the market for DIA, QQQ, and SPY
when Island electronic communications network ceased displaying its limit order book in
response to SEC’s regulatory ATS. They find that overall market quality is impaired—
Island’s market share and price discovery decreased, fragmenting the market.
As described above, there are numerous studies related to multimarket trading.
However, the results of these studies are inconsistent. Our study will shed light on this
inconsistency by examining both the short-term and long-term effects of multimarket
trading. We analyze trading costs on the existing markets to see if there is any change
when new markets enter. Hamilton (1979) advances two hypotheses regarding the
impacts of off-exchange trading on trading costs of the primary exchange. The first is the
competition hypothesis, which predicts that competition from other markets will prompt
the existing market to narrow spreads in an attempt to retain its trading volume. The
second is the fragmentation hypothesis, which suggests that the existing market will
widen spreads in order to compensate for higher costs associated with the declining
economies of scale due to lower trading volume on that market. This issue is also
addressed by Amihud and Mendelson (1996), who argue that a fragmented market
system may lead to higher spreads because of reduced liquidity. When an order of a
given size is split among market centers, there are fewer limit orders or quotes available
on the other side at any given price, likely causing the spreads to widen. We test whether
trading costs on the existing markets decline (Competition Hypothesis) or increase
(Fragmentation Hypothesis) after a new market enters.
Besides trading activity and trading costs, we also examine the impact of market
fragmentation on price impact, price volatility, and price efficiency. Price impact may
also change when a new market enters. According to Pagano (1989), liquidity can be
measured by how readily a market can absorb a trade without an adverse price change.
Thus, the price impact of a trade is a measure of liquidity. Amihud and Mendelson
(1996) argue that, in a multiple market trading environment where order flow is split
among market centers that are not perfectly coordinated, price impact may increase
because an order of a given size has fewer limit orders or quotes available on the other
side at any given price. In contrast to Amihud and Mendelson (1996), we suggest that
multimarket trading may decrease price impact because traders can split their trades
among trading venues to reduce the amount of information they reveal through their
Several studies suggest that price volatility in a consolidated market differs from
that in a fragmented market. Mendelson (1987) suggests that fragmentation increases
price variance in each separate fragment, but tends to reduce the overall variance.
There are two opposite effects on price variance when a market becomes fragmented. First, when a
market is fragmented, order intensity in each submarket declines, consequently price variance in each
separate fragment increases. Price variance within each submarket is an increasing, but concave, function
of the number of fragments. Second, when computing the average of the fragmented market prices, there is
an effect of price diversification across submarkets. Diversification is an inverse linear function of the
number of fragments. The diversification effect dominates and causes the weighted-average price variance
to be a decreasing function of the number of fragments.
contrast, Madhavan (1995) suggests that price volatility, measured by the expected
absolute change in price, is higher in a fragmented market than in a consolidated market.
In addition, Bennett and Wei (2003) find that price volatility declines when firms move
from NASDAQ to the NYSE, and attribute this improvement to the consolidation effect.
Price efficiency reflects the speed of price adjustment to new information.
Amihud and Mendelson (1996) suggest that the price discovery process is more difficult
as the trading becomes more dispersed. Since price discovery and price efficiency are
directly related, price efficiency may deteriorate when a market becomes dispersed. This
view is shared by Alan Greenspan who states that the price formation process may be
3. Data and methodology
Our data consists of all trades and quotes for DIA, SPY, and QQQ from the time
they were initially listed on the AMEX until the end of 2002. Trade and quote data is
from the New York Stock Exchange Trade and Quote database (TAQ). As TAQ does not
provide information about trade direction, we use the Lee and Ready (1991) algorithm to
identify whether a trade is buyer- or seller-initiated. We incorporate a five-second delay
as suggested by Lee and Ready (1991). A trade is classified as buyer-initiated if the trade
price is greater than the prevailing National Best Bid and Offer (NBBO) midpoint, and
seller-initiated if less than the prevailing quote midpoint. Trades executed at the
midpoint on an up-tick are classified as buyer-initiated and on a downtick as seller-
initiated. The NBBO is inferred from TAQ. When calculating the NBBO, we exclude
In a speech made by Alan Greenspan before the 2000 Financial Markets Conference of the Federal
Reserve Bank of Atlanta held in Sea Island, Georgia, October 16, 2000.
stale quotes of more than 10 minutes. Additionally, in order to avoid data errors, we
exclude all observations with recorded quoted/effective spreads greater than $3. We also
exclude all trades and quotes that occur prior to 9:30 A.M. or after 4:00 P.M. and that are
flagged as out of sequence or contain errors.
We measure trading costs using effective spread and realized spread. The
effective spread takes into account that trading occurs at prices inside the posted bid and
ask quotes. Investors frequently receive improved prices and thus encounter spreads
lower than the quoted ones. We define the effective spread at time t as:
i,t trade i,t
(Ask +Bid )
ES = 2*I * Price -
is effective spread at time t for security i, I
is a trade direction
indicator that equals one for buy orders and negative one for sell orders. We also
compute the realized spread as suggested by Huang and Stoll (1996). They decompose
the effective spread into an adverse information component and a component realized by
the liquidity providers. Realized spread is the revenue to liquidity providers. Realized
spread is defined as:
i,t trade i,t i,t+10
RS =2*I *(Price -M ) (2)
is realized spread at time t for security i, and M
is the quote midpoint 10
minutes after the trade.
Price impact measures a decrease (increase) in the price of a security following a
sell (buy) order and is also referred to as information component as it reflects the
information content of the orders. Price impact is calculated as:
PI -=I *(M M ) (3)
is price impact for trade i occurring at time t, and M
is quote midpoint
at the time of the trade.
To examine the long-term effects of multimarket trading, we apply a widely used
measure of market concentration, the Herfindahl-Hirschman Index (HHI). The HHI is
HHI = S
S is market share of a trading venue. The HHI is the sum of the squares of the
market shares of all markets. The HHI is influenced both by the number of markets and
differences in their relative size. The value of the HHI decreases as the number of
markets rises. Similarly, the value of the HHI increases when the degree of inequality in
market sizes is larger. In this study we use the HHI as a proxy for trading concentration.
Trading concentration is the opposite of market fragmentation or trading dispersion. The
HHI decreases as the trading is more dispersed.
4. Empirical analyses and results
4.1. Descriptive statistics
Table 1 reports market share for all market centers on an annual basis. All three
ETFs were primarily traded on the AMEX immediately after listing. The AMEX is
losing market share gradually over the years as trading off the primary exchange
increases. However, trading off the primary exchange is not prevalent until 2001 and
2002 when AMEX’s market share drops dramatically and it is no longer the dominant
market. The proliferation of electronic communication networks (ECNs) contributes to
this change. The increasing presence of ECNs is primarily a result of regulatory changes
brought about by the implicit collusion scandal involving market makers in the mid
1990s (Christie and Schultz, 1994). Most ECNs report trades to the NASDAQ in our
study period. Consequently, we are unable to separate ECN trades from market-maker
trades. Table 1 shows that NASDAQ’s market share increases dramatically in 2001 and
2002. Similarly, trades on the NSX and PAC also increase substantially from year 2001
to 2002. These increases are due to Island reporting trades in ETFs to the NSX and to the
partnership of Archipelago ECN and the PAC in 2002.
4.2. The short-term effects of multimarket trading
Trading off the primary exchange did not happen instantaneously as the new
markets did not enter concurrently. In this section, we examine the impact of each entry
on market share and execution quality of the existing markets, as well as of the market as
4.2.1. Change in market share
We test the liquidity externalities hypothesis, which asserts that it is difficult for a
new market to gain market share from the incumbent market as traders tend to gather in a
marketplace where there are more traders (Domowitz, 1995 and Domowitz and Steil,
1999). Bringing traders together creates liquidity externalities as the additional traders
arriving in the marketplace reduce trading costs and search costs for all investors.
After March 2002, some of Island trades in ETFs are printed in TAQ as NSX trades and some as 3
market trades. Prior, Island trades are printed as 3
market trades. In 2000, Archipelago ECN partnered
with the Pacific Stock Exchange to create a fully electronic stock exchange, the Archipelago Stock
Exchange. The rollover of ETFs to the new exchange platform occurred in August 2002. Before,
Archipelago ECN trades in ETFs are reported in TAQ as 3
market trades. Trades from market makers
and other ECNs are printed as 3
According to the
Liquidity Externalities Hypothesis, we expect new markets to divert
more order flow from existing secondary markets than from the primary market.
An alternative suggestion is that a new market that has a niche will successfully
gain order flow. We expect that the new market will emphasize the services in which it
has a comparative advantage. Consequently, we expect that the new market will target
existing markets with similar size. We refer to this conjecture as the
Advantage Hypothesis. For instance, the AMEX, NYSE, and NASDAQ compete with
each other and regional exchanges compete with each other. Battalio (1997) shows that
third market broker-dealers diverted more volume from the regional stock exchanges than
from the NYSE. Boehmer and Boehmer (2003) document that when the NYSE enters
the market for the QQQ, SPY, and DIA, trading volume is diverted from the AMEX,
NASDAQ, and regional exchanges - declines of 10 percent, 6 percent and 2 percent,
respectively. These studies lend support to the argument that new markets compete
directly with markets of similar size.
Table 2 presents the market share for each market center before and after a new
market enters. Pre (Post) includes five days before (after) a market entrance. The
minimum length of time between two entries is five days. Using a non-overlapping five-
day period enables us to separate the effects of each entry. The results from Table 2
coupled with the results from Table 1 indicate that the new entrants overcome the
liquidity externalities of the primary market, as the market share of the primary market
We also examine whether the new market competes with markets of similar size.
Each time a new market enters, we take note of the market share of each market. There
are very few large scale entries for the three ETFs. As shown in Table 2, the NYSE takes
27.77 percent and the PAC takes 2.79 percent of DIA market share upon entry. The
entrance of the other exchanges gains one percent or less of the DIA market share. The
markets that obtain much more than one percent of the QQQ market share upon entry are
the CHX (7.26%) and the NYSE/CBOE (12.11%). For SPY, only CHX (4.47%) and the
NYSE (21.26%) take more than one percent of market share immediately after entry. A
close look at the competing markets indicates no clear pattern. For a large scale entry,
such as the NYSE, all existing markets lose market share. However, the results are
ambiguous for small scale entries. We cannot, therefore, definitively conclude if order
flow is diverted from markets with similar sizes.
4.2.2. Change in trading costs
We analyze the impact of each new entry on trading costs of each existing market
as well as of the market as a whole. Table 3 reports the effective spread for each trading
venue before and after a new market entry. For the DIA, the entry of the NYSE and PAC
significantly reduces effective spread for almost all existing trading venues. Boehmer
and Boehmer (2003) argue that the decline in trading costs after the NYSE entry proves
there is a lack of competition at the time of the NYSE entry. Consistent with Boehmer
and Boehmer, our study shows that the NYSE entry makes the market for DIA more
competitive. However, competition does not lead to market equilibrium, since the entry
of the PAC one year later further reduces market-maker rents. The entrance of the CHX
into the DIA market leads to an increase in spread. The entrances of the NSX and
NYSE/CBOE in the QQQ market result in lower trading costs. It is unclear how the
entrance of the CHX into the DIA and the NSX into the QQQ can have a significant
impact given their small market share. The entrances of NSX and NYSE into the SPY
market also cause effective spreads on existing markets to decline while the entrances of
CHX and CBOE lead to increases in spreads.
The overall market quality in terms of trading costs depends on the competition
and fragmentation effects. As noted in section 2, both theoretical and empirical studies
offer conflicting predictions about this dimension of execution quality. In this study, we
argue that the change in overall transaction costs depends on whether a new market is
successful in gaining market share. A market center that attracts a large fraction of order
flow will have a greater impact on competition and fragmentation than a market center
that obtains a small proportion of order flow. This is supported by Wahal (1997) who
analyzes the impact of entry and exit of market makers on spreads and shows that a large-
scale exit (entry) is associated with a substantial decline (increase) in spreads.
We compute the changes in the overall trading costs before and after each
entrance. Table 3 shows that large scale entries, mainly the NYSE, lead to reductions in
the overall trading costs while small scale entries have ambiguous effects on the trading
We also compute the change in the overall trading costs as measured by the
percentage effective spreads and realized spreads both in percentage and in dollars (not
tabulated). The results are similar to these reported in Table 3.
In order to directly relate the change in trading costs to the market share of the new entrant we run a
regression of change in spread on changes in the determinants of spreads and market share of the entrant.
We find that the change in trading costs is significant and negatively related to the market share of the
entrant (not tabulated).
4.3. The long-term effects of multimarket trading
In this section we examine volatility, price efficiency, price impact, and trading
cost over time as trading becomes more dispersed. Previous studies provide controversial
results regarding the effects of multimarket trading. We expect the effect of multimarket
trading will be more apparent as the market of the ETFs evolves over time. A long-term
study helps determine whether the competition or the fragmentation effect dominates.
4.3.1. Price impact, price volatility, and price efficiency
Table 4 presents the correlation between the HHI and the percentage of trading
off the primary exchange and between the HHI and the number of markets. Both
variables are highly correlated with the HHI. We conclude with confidence that the HHI
is a good proxy for market concentration (opposite of trading dispersion). Additionally,
to determine the relation between market fragmentation and price impact, we compute
the correlation between HHI and price impact. The results in Table 4 show that
fragmentation is associated with lower price impact for QQQ. Opposite results are
obtained for DIA and SPY.
We also analyze the impact of market fragmentation on price volatility. We
analyze transitory volatility in our ETF time-series by looking at the standard deviation of
5-minute returns. We compute the return using the quote midpoint rather than trade price
since Bennett and Wei (2003) report that using trade prices to compute returns gives an
upward bias due to bid-ask bounce. The correlation analysis in Table 4 shows that
volatility decreases as the market become more fragmented.
We use the autocorrelation of 5-minute returns as a proxy for price efficiency.
Autocorrelation is widely used for measuring the dependence of successive price changes
in order to test the market efficiency hypothesis. In Table 4 we compute the correlation
autocorrelation and HHI. Since an autocorrelation close to zero indicates a
higher level of price efficiency, we use the absolute value of autocorrelation in our
correlation analysis. A positive and significant correlation between
autocorrelation for all three ETFs indicates that autocorrelation is smaller when the
degree of fragmentation is greater. This result is inconsistent with the view that
fragmentation harms price efficiency.
We relate volatility to HHI, trading activity variables, and rule change variables in
a multivariate analysis. The regression results in Table 5 Panel A do not support the
hypothesis that volatility is associated with market fragmentation, as the coefficients of
HHI are not significant. For the DIA and QQQ, the variation in volatility is explained
by the variations in price, number of trades,
MKT+ (the CRSP equally-weighted return if
the return is non-negative and zero otherwise), and decimalization. The coefficient of
MKT+ is negative and significant for DIA and QQQ, indicating that a rising
market is associated with lower return volatility. For the rule change variables,
decimalization is associated with lower volatility for DIA and QQQ. Coefficients of the
variables DEC and 1/64
are not significant for SPY. Overall, the results are not
consistent with Madhavan (1995) and Bennett and Wei (2003) who suggest that volatility
increases when the market becomes more fragmented.
We also examine the relationship between price efficiency and the degree of
market fragmentation in a multivariate analysis. In addition, we also relate market
transparency to price efficiency. Price efficiency may improve as the market becomes
more transparent (Chowdhry and Nanda, 1991; Madhavan, 1995; Huang, 2002). In 1996,
the SEC adopted the OHRs in an attempt to make the market more transparent. We
determine if the OHRs contribute to the improvement of price efficiency and mitigate the
negative effect of market fragmentation. We regress the proxy for price efficiency on the
HHI and an OHR dummy variable. The regression applies only to SPY, because DIA
and QQQ were introduced after the implementation of the OHRs.
OHR is a dummy
variable that equals one for the time period after the implementation of the order handling
rules and zero otherwise. The result of the regression shows that fragmentation is
associated with a higher degree of price efficiency (Table 5 Panel B). Additionally, price
efficiency improves after the implementation of the OHRs.
4.3.2. Trading costs
In this section we directly test the competition hypothesis against the
fragmentation hypothesis. The correlation analysis (Table 4) provides a univariate
analysis for the relation between HHI and trading costs. We observe a positive and
significant correlation between HHI and spread for DIA and QQQ, meaning that, as
trading becomes more dispersed, trading costs decline. This result holds for both
effective and quoted spread. Contrary to this result, the correlation between HHI and
spread is negative for SPY. The negative relationship between the HHI and spread for
SPY may be caused by the OHRs. We believe the OHRs may affect the sensitivity of
spread to fragmentation. We suggest that market transparency is an important factor that
may determine the outcomes of the competition and fragmentation effects. Chowdhry
and Nanda (1991) argue that information asymmetry increases when a market is
fragmented because informed traders can split their trades across trading venues in order
to reduce the amount of information they reveal through their trades. Increases in
information asymmetry may induce wider spreads. Related to the transparency issue, on
September 23, 2002, the Island complied with the SEC’s Regulation ATS and chose to
stop displaying its limit order book for the DIA, QQQ, and SPY. Consequently the
market for these ETFs became fragmented. Hendershott and Jones (2005) find that
overall market quality is impaired, in particular, trading costs increase and price
discovery falls during the period after Island ceased displaying its limit order book.
Since the end of the nineties, the SEC has strived to make the national market
system more transparent by imposing the limit order display rule. Consequently, the
OHRs may mitigate information asymmetry that is associated with fragmentation. We
divide the data for SPY into two sub samples with the beginning of 1997 as the cut-off
point. The choice of this cut-off point is motivated by the regulation changes that
occurred at that time. In the pre-OHRs sample, the correlation between the HHI and
spread is negative while it is positive in the post-OHRs sample (not tabulated). The result
of the post-OHRs is similar to the results of the DIA and QQQ. This phenomenon is
further investigated in the multivariate analysis by including an interaction term to allow
different slopes of the variable “HHI” in the pre- and post-OHRs periods.
To check the robustness of the decline in spreads over time as the trading
becomes more dispersed we use a multivariate analysis where we control for the changes
in trade characteristics, tick sizes, and other factors known to affect the spread, such as
overall market liquidity, and the OHRs.
12 3 4 6
8 9 10 11
Spread=α+β TSize+β (1/Price)+β Log(Vol)+β Volatility+β HHI+β MKT+
+ββDec+β (1/64th)+β OHRs+β HHI*OHRs+ε
- Spread is either the effective or quoted spread
- TSize is the mean trade size
- Vol is the average daily volume in shares
- Volatility is computed as the standard deviation of the 5-minute returns using the
- HHI is the Herfindahl-Hirschman Index
- MKT+ is the CRSP equally-weighted return if it is non-negative and zero otherwise.
- CrspVol is the aggregate dollar volume for all securities listed in the CRSP database.
This variable is a proxy for the overall market liquidity.
- Dec is a binary variable that equals one if in a decimalized environment and zero
is a binary variable that equals one if the stock is traded with a $1/64
and zero otherwise.
- OHRs is a binary variable that equals one if in the period subsequent to the OHRs
implementation and zero otherwise.
Tables 6 presents the results of the multivariate regression for quoted spreads.
We control for the determinants of spread (trade size, price, volume, and volatility) and
overall liquidity of the market. We also control for the tick size changes. The changes in
tick size for ETFs are different from that of other exchange listed securities. Prior to
April 11, 1994, ETFs were traded on a $1/32
tick. The tick size changed to $1/64
April 11, 1994 and to decimals on January 29, 2001. Harris (1994) predicts that a smaller
tick size will lead to narrower spreads. The minimum tick size limits the minimum bid-
ask spread that can be quoted. Additionally, in a market system with price-time priority,
the tick size represents the cost of obtaining price priority. With a smaller tick size, it is
less costly for market makers to step in front of the existing NBBO. The lower costs of
stepping ahead give more incentives to market makers to step ahead than if the tick size
were greater. The action of stepping ahead causes the spreads to narrow. Empirical
evidence supporting this statement includes Goldstein and Kavajecz (2000), Jones and
Lipson (2000), and Chung and Van Ness (2001). The reduction in spread is expected to
be greater for actively traded securities since their spreads are usually narrower and more
likely to be affected by the spread width constraint. The three ETFs examined in this
study are heavily traded. Hence, we expect the tick size changes from $1/32
and then to decimals to lead to narrower spreads. The results in Table 6 show that the
changes from $1/32nd to $1/64
and from $1/64
to decimals are associated with
declines in the spread. The results hold for all three ETFs and both the effective and
In order to assess the impact of market movement on spread, we include the
MKT+. Results indicate that a rising market is associated with a lower spread.
This result is consistent with Chordia, Roll, and Subrahmanyam (2001) and Van Ness,
Van Ness, and Warr (2005).
In 1996, the SEC imposed OHRs on all U.S. markets in response to the concerns
about the anticompetitive activities of NASDAQ market makers. Although the OHRs
were intended to make changes in the operation of NASDAQ markets, they were also
initiated to foster the development of an efficient, competitive, and transparent national
market system. Several studies document that market quality for NASDAQ-listed
securities significantly improves after the OHRs (Barclay, Christie, Harris, Kandel, and
Schultz, 1999; Weston, 2000; Bessembinder, 1999). Odders-White (2004), the only
study to examine the impact of the OHRs on NYSE-listed securities trading on
NASDAQ, finds that both quote quality and quoting frequency diminish for NYSE-listed
securities traded on NASDAQ. The implementation of the OHRs for exchange-listed
securities is January 10, 1997. Table 6 indicates that the dummy variable
significantly negative. However, the dummy coefficient does not represent the effect of
OHRs as we allow the slope coefficients to change. To assess the effect of OHRs on
spread, we examine the marginal effect evaluated at the mean of
HHI. The marginal
effects are significant at the one percent level (not reported). This result implies that the
spread is declining in the post-OHRs period.
The coefficients of the
HHI are positive and significant for the DIA and QQQ.
Hence, the positive relation between the
HHI and spreads is robust after controlling for
the determinants of spread and rule changes. However, the coefficient of HHI is not
significant for the SPY. Our univariate analysis indicates that the relation between HHI
and spreads may be affected by the OHRs. To investigate this issue, we include an
interaction term (
HHI*OHRs) in the spread regression. The interaction terms allow
different slopes for the Pre- and Post-OHRs periods. The results show that the interaction
terms are significant. The coefficient of the variable “HHI” measures the sensitivity of
spread to fragmentation before the OHRs. An insignificant coefficient indicates that
spread is not related to fragmentation in the period before the OHRs. The sum of the
coefficient of HHI and the corresponding interactive variable represents the sensitivity of
spread to fragmentation after the OHRs. To assess whether the sensitivity of the spread
to fragmentation is significant in the post OHRs period, we compute a t-ratio given that
OHRs is equal to one. We find that the t-ratio is 13.40 (not tabulated). Given the
positive and significant coefficients for the variable HHI in the post-OHRs period, we
conclude that fragmentation is associated with lower spreads in this period. This result is
consistent with our prediction that the sensitivity of spread to fragmentation is affected by
the OHRs. We conclude that trading costs for these three ETFs decline as trading
becomes more dispersed
. The results offer evidence to support the competition
In order to assess the impact of trading off the primary exchange on trading costs,
HHI with OFF in the second regression. OFF is the percentage of share
volume executed off the primary exchange. The results show that the percentage of
trading off the primary exchange is inversely related with trading costs for DIA and SPY.
These results do not hold for the QQQ. It is worth noting that OFF is not a proxy for
trading concentration. Trading concentration is affected by both the number of markets
and the degree of inequality in market share of these markets. Market concentration can
be high when off exchange trading gravitates to a single venue. We also include an
interaction term OFF*OHRs for the SPY regression. The results show that spread is
negatively related to off exchange trading in the pre-OHRs period but positive in the
Overall, the results of the multivariate analysis offer evidence supporting the
competition hypothesis. Multimarket trading is not associated with spread in the pre-
OHRs period. However, in the post-OHRs period, multimarket trading is associated with
lower spreads. We show that the OHRs significantly improve overall market quality and
mitigate fragmentation effects that may have resulted from the dispersal of trading. We
We acknowledge that improvement in market quality, i.e., a decline in trading costs, may be due to the
primary market being a poor market, so that when the primary market loses market share, market quality
improves. Goldstein, Shkilko, Van Ness, and Van Ness (2006) show that the AMEX has 0.08 percent of
trading volume and the highest trading costs for the NASDAQ 100 stocks. The AMEX has lost market
share in NASDAQ and listed stocks, but has remained in business by creating the ETFs. However, it is
also beginning to lose market share in ETFs. We would like to thank an anonymous referee for pointing
out this possible explanation.
also run regressions of effective spreads. The results are similar to those of the quoted
spread regressions (not tabulated).
The results of our examination of three ETFs over an extended time period show
that new markets are able to overcome the liquidity externalities enjoyed by the primary
market. The short-term effect of a new market entry depends on the size of the new
market. Large scale entries, mainly the NYSE, improve overall market quality by
reducing trading costs, while small scale entries show mixed effects.
The results of the long-term study indicate that the relation between market
fragmentation and spread is affected by the OHRs. In pre-OHRs period, spread and
fragmentation are not related. In the post-OHRs period, fragmentation is associated with
lower spread. This evidence is consistent with the view that the attempt by the SEC to
make the market more transparent helps suppress the fragmentation effect of multimarket
trading. Our analysis indicates that the competition effect dominates the fragmentation
effect over a long-term horizon as reflected in the declines in trading costs when
fragmentation increases. We also present evidence against the view that multimarket
trading harms price efficiency and increases price volatility. Our analysis provides
evidence to support Stoll (2001) who states that the extent of market fragmentation and
its adverse effects are probably overstated.
Amihud, Y. and H. Mendelson, 1996. A new approach to the regulation of trading across
71 New York University Law Review 1411.
Arnold, T., P. Hersch, J. H. Mulherin, and J. Netter, 1999. Merging markets,
Finance 54, 1083-1107.
Barclay, M. J., W. G. Christie, J. H. Harris, E. Kandel, and P. H. Schultz, 1999. Effects of
market reform on the trading costs and depths of Nasdaq stocks,
Journal of Finance 54,
Battalio, R. H., 1997. Third market broker-dealers: Cost competitors or cream
Journal of Finance 52, 341-352.
Bennett, P. and L. Wei, 2003. Market structure, fragmentation and market quality -
Evidence from recent listing switches,
Working paper, New York Stock Exchange.
Bernhardt, D. and E. Hughson, 1997. Splitting orders,
Review of Financial Studies 10,
Bessembinder, H., 1999. Trade execution costs on NASDAQ and the NYSE: A post-
Journal of Financial and Quantitative Analysis 34, 387-407.
Boehmer, B. and E. Boehmer, 2003. Trading your neighbor's ETFs: Competition or
Journal of Banking & Finance 27, 1667-1703.
Chordia, T., R. Roll, and A. Subrahmanyam, 2001. Market liquidity and trading activity,
Journal of Finance 56, 501-530.
Chowdhry, B. and V. Nanda, 1991. Multimarket trading and market liquidity,
Financial Studies 4, 483-511.
Christie, W. G. and P. H. Schultz, 1994. Why do NASDAQ Market Makers Avoid Odd-
Journal of Finance 49, 1813-1840.
Chung, K. H. and R. A. Van Ness, 2001. Order handling rules, tick size, and the intraday
pattern of bid-ask spreads for Nasdaq stocks,
Journal of Financial Markets 4, 143-161.
De Fontnouvelle, P., R. P. H. Fishe, and J. H. Harris, 2003. The behavior of bid-ask
spreads and volume in options markets during the competition for listings in 1999,
Journal of Finance 58, 2437-2463.
Domowitz, I., 1995. Electronic Derivatives Exchanges: Implicit Mergers, Network
Externalities, and Standardization,
Quarterly Review of Economics and Finance 35, 163-
Domowitz, I. and B. Steil, 1999. Automation, trading costs, and the structure of the
securities trading industry,
Brookings-Wharton Papers on Financial Services, 33-82.
Goldstein, M. A. and K. Kavajecz, 2000. Eights, sixteenths and market depth: changes in
tick size and liquidity provision on the NYSE,
Journal of Financial Economics 56.
Goldstein, M. A., A. Shkilko, B. F. Van Ness, and R. A. Van Ness, 2006, Competition in
the market for NASDAQ-listed securities, working paper.
Hamilton, J. L., 1979. Marketplace fragmentation, competition, and the efficiency of the
Journal of Finance 34, 171-187.
Harris, L., 1993. Consolidation, fragmentation, segmentation and regulation,
Markets, Institutions & Instruments
Harris, L., 1994. Minimum price variations, discrete bid-ask spreads, and quotation sizes,
Review of Financial Studies 7, 149-178.
Hendershott, T. and C. M. Jones, 2005. Island goes dark: transparency, fragmentation,
Review of Financial Studies 18, 743-793.
Hendershott, T. and H. Mendelson, 2000. Crossing networks and dealer markets:
competition and performance,
Journal of Finance 55, 2071-2115.
Huang, R. D., 2002. The quality of ECN and Nasdaq market maker quotes,
Finance 57, 1285 - 1319.
Huang, R. D. and H. R. Stoll, 1996. Dealer versus auction markets: A paired comparison
of execution costs on NASDAQ and the NYSE,
Journal of Financial Economics 41, 313-
Jones, C. M. and M. L. Lipson, 2000. Sixteenths: Direct evidence on institutional trading
Journal of Financial Economics.
Lee, C. and M. Ready, 1991. Inferring trade direction from intraday data,
Macey, J. R. and M. O'Hara, 1999. Regulating exchanges and alternative trading systems:
a law and economic perspective,
28 Journal of Legal Studies 17.
Madhavan, A., 1995. Consolidation, fragmentation, and the disclosure of trading
Review of Financial Studies 8, 579-603.
Mendelson, H., 1987. Consolidation, fragmentation, and market performance,
Financial and Quantitative Analysis
Odders-White, E. R., 2004. Third Market Reforms: The Overlooked Goal of the SEC's
Order Handling Rules,
Journal of Financial and Quantitative Analysis 39, 277-305.
Pagano, M., 1989. Trading volume and asset liquidity,
The Quarterly Journal of
Economics 104, 255-274.
Parlour, C. A. and D. J. Seppi, 2003. Liquidity-based competition for order flow,
of Financial Studies
Stoll, H. R., 2001. Market fragmentation,
Financial Analysts Journal 57, 16.
Van Ness, R. A., B. F. Van Ness, and R. S. Warr, 2005. Nasdaq Trading and Trading
The Financial Review 40, 281-304.
Wahal, S., 1997. Entry, exit, market makers and the bid-ask spread,
Review of Financial
Studies 10, 871-901.
Weston, J. P., 2000. Competition on the Nasdaq and the impact of recent market reforms,
Journal of Finance 55, 2565-2598.
Table 1: Market Share and Aggregate Volume on a yearly basis
This table shows yearly market share and volume for DIA, QQQ and SPY. AMEX is the American Stock
Exchange, NASDAQ is the National Association of Securities Dealers, BOS is the Boston Stock Exchange,
CHX is the Chicago Stock Exchange, NSX is the National Stock Exchange, NYSE is the New York Stock
Exchange, and PAC is the Pacific Stock Exchange (now NYSE Arca).
Market Share (in terms of share volume)
ETF 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
0.9305 0.9064 0.8802 0.6359 0.3039
0.0695 0.0936 0.1198 0.2409 0.3775
0.8627 0.7140 0.4140 0.1832
0.0747 0.1206 0.3841 0.5373
0.0484 0.1247 0.0850 0.0482
0.0078 0.0241 0.0264 0.0392
0.0064 0.0090 0.0011 0.0169
0.0074 0.0268 0.0850
0.0003 0.0054 0.0108
0.9903 0.9918 0.9812 0.9726 0.9557 0.9353 0.8188 0.7814 0.5986 0.2971
0.0097 0.0083 0.0188 0.0274 0.0336 0.0225 0.1109 0.1465 0.2460 0.4641
0.0002 0.0092 0.0114 0.0047 0.0004 0.0253
0.0101 0.0291 0.0410 0.0460 0.0374 0.0175
0.0005 0.0033 0.0023 0.0068 0.0140 0.0660
0.0007 0.0156 0.0147 0.0199 0.0311
Table 2: Market Share Before and After a New Entry
This table shows the impact of each new entry on the market share for each of the ETF market centers. Pre
(Post) is five trading days before (after) an entry. Market share is computed based on the aggregate share
volume of five days before and after an entry. AMEX is the American Stock Exchange, NASDAQ is the
National Association of Securities Dealers, BOS is the Boston Stock Exchange, CHX is the Chicago Stock
Exchange, NSX is the National Stock Exchange, NYSE is the New York Stock Exchange, and PAC is the
Pacific Stock Exchange (now NYSE Arca).
AMEX NASDAQ BOS CHX NSX NYSE PAC
BOS Pre 0.7423 0.2577 0.0000
Post 0.7453 0.2532 0.0015
Diff. 0.0030 -0.0045 0.0015
CHX Pre 0.7517 0.2396 0.0088 0.0000
Post 0.7308 0.2444 0.0186 0.0062
Diff. -0.0209 0.0050 0.01* 0.0062
NSX Pre 0.7222 0.2080 0.0231 0.0467 0.0000
Post 0.7610 0.1925 0.0280 0.0079 0.0106
Diff. 0.0390 -0.0155 0.0050 -0.0388 0.0106
NYSE Pre 0.6602 0.2909 0.0163 0.0202 0.0123 0.0000
Post 0.5041 0.1799 0.0134 0.0162 0.0088 0.2777
Diff. -0.1561*** -0.1111*** -0.0028 -0.0041 -0.0036 0.2777
PAC Pre 0.2507 0.4436 0.0193 0.0490 0.1788 0.0586 0.0000
Post 0.2364 0.4237 0.0221 0.0564 0.1813 0.0522 0.0279
Diff. -0.0143 -0.0199 0.0030 0.007* 0.0020 -0.0064 0.0279
AMEX NASDAQ CHX BOS PAC NSX PHL NYSE/
CHX Pre 0.9225 0.0775 0.0000
Post 0.8256 0.1018 0.0726
Diff. -0.0969** 0.0240 0.0726