Download internalization and market quality: an empirical

INTERNALIZATION AND MARKET QUALITY:
AN EMPIRICAL INVESTIGATION
Norris Larrymore
Quinnipiac University
Albert J Murphy
SUNY College at Old Westbury
JEL Classification: G10; G11; G14; G24
We thank Amber Anand, Tamisan Agbeyegbe, Alexander Kurov, Joseph Onochie, Daniel Weaver, Gwendolyn
Webb, the FMA 2002, and the SFA 2006 Annual Meeting participants for their guidance and comments. We are
responsible for all remaining errors.
2
Abstract
We examine the impact on market quality of a rule change designed to eliminate the perceived
deleterious effects of internalization. In particular, the Toronto Stock Exchange (TSE),
concerned about a possible deterioration of liquidity on its central market, enacted the Price
Improvement Rule on October 26, 1998. We find that the rule’s introduction is accompanied by
a significant improvement in market quality. We also document a strong, negative, association
between internalization rates and spreads, suggesting that the practice of internalization is not
necessarily associated with poor market quality. Thus, a judicious regime of regulated
internalization may be associated with superior market quality.
3
I. Introduction
The worldwide, technology-driven proliferation of trading systems reflects continuing
efforts by national regulators such as the Securities and Exchange Commission (SEC) in the US
to promote competition among constituent market centers and order matching systems.1 That
proliferation has also raised concerns about its potentially harmful effects on financial asset
liquidity. Recently, major financial centers such as the New York Stock Exchange (NYSE) and
the Toronto Stock Exchange (TSE) have regulation regimes designed to mitigate those perceived
effects.
A practice associated with the order-routing decisions of brokers, the internalization of
trades, has caught the attention of regulators. Orders may be executed in-house against a firm’s
inventory, or a broker-dealer firm may route orders from its brokerage section to its market
making section, internalizing trades. Internalized order flow forms market fragments. Order
matching takes place within a fragment, which is isolated from interaction with the wider market.
Such practices can result in the violation of price and time priority precedence rules. Moreover, a
member firm’s role as the client’s agent directly conflicts with its role as principal.
We explore the impact of an attempt by the Toronto Stock Exchange (TSE) to improve
market quality by reducing the perceived negative effects of internalization and the resulting
internal market fragmentation. For example, some of the TSE’s key stakeholders voiced
concerns about its larger members’ increasing control over their trading environment, and the
potential threat to trade execution quality. Concern focused on such members’ capacity (using
their increasingly sophisticated order management systems) to efficiently route, manage, and
1
For example, the New York Stock Exchange (NYSE) was required to modify its Rule 394, which prohibited
members from engaging in the market making and trading of securities off the exchange floor, to become Rule 390,
which allowed agency transactions away from the exchange floor. In 1980, the S.E.C. instituted its own rule, Rule
4
process retail order flow from branch offices.2 While order execution speed improved, members
could effectively internalize trades without exposing orders in the book.
Specifically, on
October 26, 1998 the TSE implemented a rule, referred to as the Price Improvement Rule, or the
Customer-Principal Trading Rule, requiring all orders of 5,000 shares or less either to receive
price improvement or to be routed to the limit-order book for execution, unless the
internalization can be justified by the market. If an intentionally internalized trade is ‘justified by
the market’ then the trader must provide evidence that some or all of the following factors
favored the trade: previous trades, quoted size, size imbalances on either side of the market,
current spread and its relation to the normal spread, price trend, volume trend, other markets, and
general market activity.
Intentionally internalized trades may take place ‘on the book’ or ‘off the book’. The
former kind of internalization occurs when members, while providing speedier execution through
the use of their increasingly sophisticated order management systems, also effectively ‘hold up’
tradable orders, enter offsetting orders in the book, then release held up orders timed to cross
with offsetting orders. Such offsetting orders may match the best possible price available across
public markets, but may not necessarily improve on that price.3 The latter kind of internalization
refers to the in-house matching of client orders or the crossing of a client’s order against the
member’s inventory without exposing the client’s order in a public order book. The new rule
requires an improvement in price to the client beyond the best public price available.
19c-3, that allowed NYSE members to not only engage in off-board transactions, but to make off-board markets for
stocks that were listed on the exchange after April 26, 1979, in direct competition with the NYSE specialists.
2
3
See Toronto Stock Exchange (TSE) Special Committee Report (1997).
See Toronto Stock Exchange (TSE) Special Committee Report (1997).
5
We develop tests to assess the impact on market quality of this attempt to consolidate
order-flow on the Toronto Stock Exchange’s (TSE’s) central auction market. Market quality is
comprehensively measured along several dimensions. After the implementation of the Price
Improvement Rule, we find an increase in price improvement rates; sharp falls in both quoted
and effective spreads, especially for small orders; lower return volatility; greater market depth;
and higher overall market quality as measured by pricing error after the implementation of the
rule. Intriguingly, we also find a strong negative association between internalization rates and
effective spreads after controlling for the impact of other relevant factors, including the
introduction of the Price Improvement Rule. The unequivocal increase in market quality brought
about by the new regulation suggests that unregulated market fragmentation may lead to poor
market outcomes. However, the practice of internalization may be justified by market conditions
and can be associated with a high quality market.
II. Literature Review
Several researchers provide evidence of the harmful effects of the market fragmentation
resulting from internalizing and purchase of order flow. Chordia and Subrahmanyam (1995) and
Easley, Keifer and O’Hara (1996) surmise that order flow from relatively uninformed traders is
internalized or purchased by market makers, who then charge prices as if the order flow was
from informed traders, thus profiting by the adverse selection component of the trades. This
selective trading has the effect of raising execution costs in the central market, where order flow
then consists of a relatively high proportion of orders from informed traders, as market makers
attempt to recoup losses from informed traders by charging the adverse selection component of
6
the spread. Battalio, Green and Jennings (1998) find some evidence in support of this theory in
their examination of Merrill Lynch’s October 1995 decision to stop routinely sending orders to
Merrill-affiliated specialists on the Boston and Pacific stock exchanges. They find that relative to
a matched sample of stocks, spreads on the NYSE fall in the stocks affected by Merrill’s
decision. Bessembinder (1999) finds that execution costs on NASDAQ remain higher than those
on the NYSE after trading in both trading venues reduced their tick size and new order handling
rules were adopted on NASDAQ in 1997. He cites extensive preferencing on NASDAQ as one
possible explanation. Chakravarty and Sarkar (2002) develop a model of broker’s trading, with
applications to order flow internalization. They find that market depth and price informativeness
are lower while uninformed losses are higher. Kluger and Wyatt (2002) also model
internalization, and conclude that, in a multi-market setting, several markets reach a collusive
equilibrium with wide spreads, suggesting that the ability to internalize order flow allows dealers
to coordinate on a less competitive level.
The practices of payment for order flow and internalization could also represent
competitive, innovative responses to demands of a diverse investor clientele. Investors may seek
liquidity within decidedly different contexts, and may differ in terms of what they want most
from a trade at any given point in time, along several dimensions: obtaining a particular trade
size, anonymity, or speed may be prioritized over obtaining the best price. In the US, bid-ask
spread differences (for the same security) often occur among the (recently emerged) alternative
trading venues called Electronic Communication Networks (ECNs). The ECNs justify those
differences with the claim that they offer unique services that investors, who are fully aware of
the spread differences, may be willing to pay a premium for the unique services they offer. A
growing body of evidence suggests that internalizing and purchasing order flow may provide
7
superior trading services at lower costs. Battalio (1997) finds that effective bid-ask spreads do
not increase and quoted spreads fall when a major third market broker-dealer (Madoff Securities)
begins selectively purchasing order flow, and concludes that Madoff uses a cost advantage to
process orders. Battalio, Green and Jennings (1997) obtain results along similar lines, and
empirically demonstrate that the opportunities the Boston and Cincinnati stock exchanges offer
to internalize their orders do not have a significant effect on quoted or effective spreads. They
conclude that the presence of adverse selection is not economically significant, and that the
regional exchanges may be able to realize sufficiently strong cost efficiencies. Other studies
contend that extant empirical investigations of execution quality focus too narrowly on execution
prices. Hansch et al (1999) study execution quality, preferencing and internalization for a sample
of London Stock Exchange Stocks. They find that internalized trades pay relatively lower
spreads, a result inconsistent with the hypothesis that market-making firms collude to set
artificially high spreads through internalization. Battalio, Hatch and Jennings (2003) conclude
that consideration of nonprice dimensions of market quality such as execution speed, depth
improvement, and order flow payment can provide an assessment of market quality that is most
closely aligned with investor interests. Peterson and Sirri (2003) examine preferencing on
regional exchanges in the US, and find that while execution costs tend to be lower on the primary
market (the NYSE) than on the regionals, they are lower on the preferencing regional exchanges
relative to the non-regional exchanges. Additionally, the regionals execute limit orders more
frequently than the primary market and with no worse execution quality otherwise.
Some researchers concede that attempts to curb or eliminate practices deemed harmful to
market quality may be motivated by limited analysis, and may ignore potential benefits from
those very practices. Chung, Chuwonganant, McCormick (2004) use trade and quote data for a
8
large number of NASDAQ dealers to determine the impact of order preferencing on market
quality. While they find dealer quote aggressiveness (the spread) to be negatively (positively)
associated with the extent of internalization, they conclude that the net impact of preferencing on
market quality is difficult to determine because the practice is likely to have broad and diverse
ramifications for investor welfare. For example, order preferencing can reduce broker search
costs, allowing the savings to be passed along to customers in the form of reduced commissions.
Stoll (2001) advocates against strict, formal approaches to linking markets designed to curb the
incidence of practices that may lead to fragmented markets. Such approaches would impede
competition. Instead, a reasonable resolution should focus on providing all investors with
unfettered access to timely, comprehensive information, and the implementation of a judicious
regulatory regime that ensures that the order routing decisions of brokers are in alignment with
the desires of their clients. Investors would then be able to make timely decisions about partaking
in trades with the most appealing features, without unfairly impinging on the welfare of other
traders.
III. Methodological Overview, Data and Descriptive Statistics
Methodological Overview
The hypotheses that follow are tested using time-stamped intraday trade and quote data
obtained from the Toronto Stock Exchange. Except where otherwise specified, a hypothesis is
tested using the value of an appropriate measure defined in a pre-event period and compared
with the measure’s value in a post-event period to determine whether a statistically significant
change occurred. Since testing for a change in market quality involves the careful assessment of
change along several dimensions, several kinds of statistical tests (described in subsequent
9
sections) of varying sophistication are conducted. This is done on a stock-by-stock basis. The
pre-event period is defined as the period from 35 trading days to 11 trading days prior to the event
date (October 26, 1998), and the post-event period is from 11 trading days to 35 trading days
afterward. The measurement period consists of the pre- and post-event periods. Following
Madhavan, Porter and Weaver (2001), 10 trading days immediately before and after the event date
are excluded from the measurement period to avoid any bias in the results due to the proximity of
the measurement period to the event date. Such biases may arise because in the days just before the
rule changes, traders may have begun routing some additional orders to the floor as a way of
preparing their order management systems. Also in the first few days after the rule change, some
traders may not have adjusted to the new rules fully.
Cross-sectional tests are conducted to determine whether any significant differences in
results exist among stocks. It is most likely, for example, the TSE rule change would affect
stocks that differ by trading volume differently. In light of this possibility, the effects of the rule
change on stocks are compared according to volume quartiles. Some comparisons are made
according to trade size ranges to capture differential effects with reference to the trade size range
specified in the price improvement rule. It is also likely that the rule change could affect
particular trade types (for example, trades occurring at the spread versus trades occurring outside
the spread, etc.) differently.
Data
The TSE Equity Trading History CD-ROM products are the primary data source used in
this study. They provide historical trade and quote data for the months of September through
December 1998. Available trade information includes the stock symbol, trade date, trade time
10
(stamped to the nearest second), number of shares traded, and buyer and seller identification.
Available quote information includes the stock symbol, quote date, quote time, and bid and ask
prices and corresponding trade sizes displayed in the TSE’s limit order book. The sample
consists of all qualifying equity trade and (accompanying) inside quote information covering two
25-trading day periods before and after the introduction of the Price Improvement Rule on
October 26, 1998. The pre-event period consists of the 25 trading days from September 3rd to
October 8th 1998, 11 days prior to the introduction of the Price Improvement Rule. The postevent period consists of the 25 trading days from November 10th, 11 days after the introduction
of the Price Improvement Rule, to December 14th, 1998.
Since open trades occur under a different protocol than those made during the rest of the
day, they are eliminated from the sample. Canceled trades, and trades with no buyer or seller
identification are eliminated. Trades with negative prices (due to data errors), quotes or bid-ask
spreads are eliminated. To avoid measurement error due to changing tick sizes, trades with prices
below C$5.00 are excluded from the sample. Trades occurring before or after the TSE’s normal
trading hours of 9:30 a.m. to 4:00 p.m. are excluded. Stocks with fewer than 50 quotes in either
the pre- or post-event period are excluded from the sample. Unit stocks, international receipts,
warrants and debentures are also excluded. Firms without at least one trade in each of the 50
trading days in the sample are excluded.
Descriptive Statistics
Table 1 gives some descriptive statistics by stock trading volume quartile in the pre-event
period. The final raw sample consists of 258 stocks involved in a total of 890,404 trades. Trading
activity is concentrated among small trades (which consist of 1,200 shares or less), which make
11
up 79 percent of all trades, while relatively large trades (greater than 5000 shares) make up only
3 percent of all trades.
IV. Hypotheses and Tests
We develop several hypotheses, each of which is tested by considering an aspect of market
quality change after the introduction of the Price Improvement Rule. The tests differ in their
degree of sophistication and focus because market quality and market quality changes are best
assessed using a multi-dimensional approach. Battalio, Hatch and Jennings (2001, 2003), for
example, conclude that ‘Failure to consider all aspects of execution quality can lead to order
decisions that may not be in investors’ best interests and to suboptimal market design.’
Price Improvement
An improved price occurs within the quoted bid-ask spread. If the rule change results in
more efficient trade execution through internalization, members could pass on the cost savings to
investors in the form of improved prices. However, if quoted spreads decline following the rule
change, then opportunities for price improvement could decline and price improvement rates
could decline in general. Thus, the issue of how the rule change affects price improvement rates
is an empirical one, and its impact can be tested by comparing the percentage of trades that
occurred within quoted spreads before and after its introduction. Paired t-tests as well as the
Wilcoxon Sign Test for paired observations are employed. The relevant null hypothesis is
H1: The rule change did not affect the proportion of trades receiving price
improvement.
12
Spreads
Traders who desire to transact immediately will pay whoever supplies the necessary
liquidity—the registered trader (the TSE’s equivalent of the NYSE’s Specialist), limit order
trader or floor broker. The round-trip cost of such a transaction is the bid-ask spread. The TSE’s
introduction of its Price Improvement Rule should cause limit order traders, attracted by the
prospect of reduced time to execution brought about by an increase of market orders directed to
the book, to converge on the auction market. Quoted bid-ask spreads fall as more limit orders
compete on price. Those members who do not route trades to the order book must improve on
price, unless market conditions justify the trade. Effective bid-ask spreads decline.
Following the specification of Stoll (1989), one measure of the cost of executing a trade
is the quoted, or dollar, spread:
Dollar Spreadit=Askit–Bidit,
(1)
where Askit and Bidit are the time-weighted average ask and bid prices for stock i in measurement
period t. The time-weighted average spread in the pre-event period is compared with that in the
post-period using a paired t-test. A related execution cost measure is the relative, or percentage
spread, which measures transaction costs per dollar of investment:
Percentage Spread it 
( Ask it  Bid it )
( Ask it  Bid it ) / 2
(2)
Trades frequently take place at prices between the spread as market makers update their information
set after posting quotes. Thus, test for changes in the following measures of actual (or effective)
execution costs are conducted (see Peterson and Fialkowski 1994):
 Ask it  Bid it 
Effective Dollar Spread it  Priceit  

2


(3)
13
Effective Percentage Spread it 
Priceit  ( Ask it  Bid it ) / 2
( Askit  Bid it ) / 2
(4)
where the spread measures are share-weighted (see Battalio 1997), and Priceit is the average
transaction price of stock i in measurement period t.4
The rule change may be one of several factors that affect spreads (see Hamilton 1979;
Stoll 1989). It is well established that spreads vary inversely with trading volume, and directly
with return volatility and price. We would expect, controlling for the influence of the abovementioned factors on spreads, a decline in trade execution costs following the rule change. We
therefore estimate a regression model of each spread measure against a dummy variable that
assumes the value 0 if the trades occurs in the pre-event period, and 1 if the trade occurs in the
post-event period; controlling for the effects of volatility, trading volume, and trade execution
prices. The relevant null hypothesis is
H2: There are no statistically significant changes in trade execution costs following
the implementation of the Price Improvement Rule.
The practice of internalization could engender a split in order flow into two parts: orders
from uninformed traders that are internalized away from the central market, while relatively
informed order flow is routed to the order book. This could result in the widening of the inside
(best bid and ask) quote in the central market, as the adverse selection component of the spread,
the part of the spread charged by the market makers to protect against losses to informed traders,
increases (see Easley, Kiefer, and O’Hara 1996). With the implementation of the Price
Improvement Rule, we would expect adverse selection to decline, as members route more
4
The effective spread is also known as the liquidity premium, and provides a per share estimate of the execution
costs of a market order (that is, the cost of liquidity).
14
uninformed orders (including internalized trades justified by market conditions) to the book. The
adverse selection component of the spread is estimated using the methodology detailed in Booth,
Lin and Sanger (1995), and a paired t-test is used to determine the nature of the change of this
component around the Price Improvement Rule. The relevant null hypothesis is
H3: There are no statistically significant changes in adverse selection component of
the effective spread following the implementation of the Price Improvement
Rule.
Market Depth
Market depth, or the volume of orders at the inside quote is an important measure of
market quality. Investors seek a liquid market that provides the maximum number of offsetting
orders at the best prices. Such competitive conditions provide the investor with cheap, timely
execution. In an auction market, market depth is provided by the entry, consolidation, and
display of limit order bid and ask prices along with their trade sizes in a central order book. Lee,
Mucklow and Ready (1993), and Harris (1994) note that market depth assessments, in addition to
bid-ask spread estimates, provide a more complete description of market quality. The
introduction of the price improvement rule should attract limit order traders (as described above)
to the auction market; as such traders would have greater confidence that their orders would
enjoy maximum price competition, resulting in greater market depth. The impact of the new rule
on market depth can be determined by comparing the time-weighted average quoted depth,
measured by the arithmetic average of the number of shares offered at the bid price, and the
number of shares offered at the ask price, in the pre- and post-event periods using a paired t-test.
15
However, Harris (1994), Chordia, Roll, and Subramanyam (2000), and Heflin and Shaw
(2001), while they find that market depth is negatively related to quoted spread and positively
related to trading volume, all find that market depth increases with trading volume. Additionally,
one would expect market participants to offer more shares (at best-quoted prices) when return
volatility is low (see Heflin and Shaw 2001). To test for such effects, market depth is regressed
against a dummy variable that assumes the value 0 if the trade (and corresponding quote)
occurred in the pre-event period, while it assumes the value 1 if the trade occurred in the postevent period. Other included explanatory variables are return volatility, quoted spread, and
trading volume in an attempt to control for the influence of those factors on market depth. The
relevant null hypothesis is
H4: There was no change in mean quoted depth following the implementation of the
Price Improvement Rule.
Volatility
Temporary, but frequent order imbalances lead to short-term price instability and volatility. The
introduction of the Price Improvement Rule would reverse that trend as the increased order flow
and return of limit order traders to the auction market would increase market depth, resulting in
shorter-lived order imbalances and more price stability. Short-term price volatility would decline
as a result. There is an established (positive, information-induced) relationship between market
volatility, and trading activity in a stock (see Easley and O’Hara 1987; Harris and Raviv 1993;
and Jones, Kaul and Lipson 1994). To control for this influence on the stock volatility, return
volatility is regressed against the number of transactions in a given stock in the pre- and postevent periods. The relevant null hypothesis is
16
H5: The implementation of the Price Improvement Rule did not affect short-term
stock price volatility.
A Measure of Overall Market Quality: The Pricing Error
The mechanism by which a market determines a price that reflects traders’ aggregate
desire to hold shares of a security is called its price discovery mechanism. In a perfectly efficient
market, any new public information about the value of a security is immediately and costlessly
impounded in its fair (or equilibrium, or efficient) price, which is equal to its transaction price,
which follows a random walk.
5
In reality, if the price discovery process is not instantaneous
prices fluctuate as the market attempts to determine a new equilibrium price following a change
in the demand to hold shares of the security. Further, inaccurate price determination may result
from traders who do not immediately convey their orders to the market, or from extant limit
orders that are not regularly revised, leading to a lagged adjustment of prices to information.
It is convenient to partition the discrepancy between the transaction price and the
equilibrium price, heretofore known as the pricing error, into two components: pricing errors due
to the inefficient and lagged translation of information into prices (that is, information- related
market impact) and other errors. Following Hasbrouck (1993), transaction price decomposition is
modeled using a general Vector AutoRegression (VAR) framework that allows serial correlation
in returns and other explanatory variables. Specifically, the logarithm of the observed transaction
price of a security is expressed as the sum of two components:
pt  mt  st
5
(5)
Hasbrouck (1993) defines the efficient price as the ‘final, end-of-trading, value of the security conditional on all
public information at time t, including whatever private information may be inferred from the published terms of the
transaction’.
17
where mt is the efficient price, defined as the expectation of the end-of-trading value of the
security conditional on all public information available at transaction time t, including whatever
private information may be inferred from the published terms of the transaction, and st is the
deviation of the transaction price from the efficient price. The efficient price is assumed to
evolve as follows:
mt  mt 1  x t  ut
(6)
where xt is the signed log of the square root of the trade size, with a negative (positive) sign if
the transaction price falls below (above) the misquote; xt represents information that the market
infers from the trade direction and size respectively, and ut is an innovation that is uncorrelated
with x t and the result of non-trade information. Note that E [ut ] = 0, E [u2t ] = 2u, and E[utu ] =
0, for t  . Similarly, the pricing error is expressed as a function of information related errors
and non-information related errors:
st  x t  ut  t
(7)
In equation (7), x captures market impact effects due to price discovery errors and other lagged
price adjustment effects, while the u term captures pricing errors due to transient market impact
effects (such as large trades, for example), inventory control mechanisms, and price discreteness.
It is assumed that st is a zero-mean, covariance-stationary process. Market quality can be
assessed by focusing on the standard deviation of the pricing error, s, which can be construed as
a summary measure of market quality, and describes how closely the transaction price tracks the
18
efficient price. A lower (higher) value of s implies (lower) higher market quality. We would
expect that the introduction of the price improvement rule to result in greater, more timely
exposure of orders in the order book, as broker-dealers become less motivated to profit at the
expense of clients traders by isolating their trades. Price discovery errors should decline, as
traders use more accurate information in forming expectations. Also, transitory price impacts
related to large trades would decline and market depth should increase, as more orders are
exposed in the order book. In general, market quality should improve after the introduction of the
Price Improvement Rule. A comparison of the standard deviation of the pricing error in the preevent period with that in the post-event period can test for a change in the standard deviation of
the pricing error. Specifically, an assessment of the effect of the rule change on the standard
deviation of the pricing error is made using the Wilcoxon one-tailed sign-ranks test (see Kumar,
Sarin, and Shastri 1998). The relevant null hypothesis is
H6: There was no change in overall market quality (as measured by the standard
deviation of the pricing error) after the introduction of the Price Improvement
Rule.
V. Findings
Price Improvement
In Table 2, we find a uniform, significant decline in price improvement rates across trade
size ranges, suggesting a decline in opportunities for price improvement, given a drop in quoted
spreads after the introduction of the Price Improvement Rule. We thus reject H1 (at 5% level of
significance), and conclude that the new rule significantly impacted price improvement rates.
19
Spreads
Tables 3 gives results of a test of the effect of the rule change on quoted spreads,
controlling for the influences of other pertinent factors. While stock price, trading volume, and
return volatility all appear to significantly impact quoted spreads (in the expected directions), the
rule change also has a significant impact, as reflected in the significant, negative dummy variable
coefficient value. This result provides stronger evidence that the rule change reduced quoted
spreads. We conjecture that limit order traders, attracted by the resulting market conditions
which protect them from exploitation by market makers, are encouraged to submit limit orders,
and in so doing contribute to greater price competition. Quoted prices improve as a result, and
the quoted bid-ask spreads fall. Brokers who previously earned abnormal profits and / or realized
cost savings through the use of more efficient trade execution systems (for example, their order
management systems) are now forced to pass those savings on to clients.
Table 4 shows the results of regression analyses for effective spreads. A significant
decline in the dollar effective spread is recorded overall; while significant, strong declines are
recorded for all considered trade size ranges. Percentage spread declines (overall and for specific
trade size ranges) are even stronger (table not shown). Thus, the rule change has achieved the
desirable result of lowering effective costs of trading by reducing opportunities for the
exploitation of retail investors by more sophisticated and better-informed market participants.
We thus reject H2 at the 5% level of significance, and conclude that the Price Improvement Rule
did have a significant impact on trade execution costs.
20
Table 5 shows that except for the low-volume, illiquid stocks, strong declines in adverse
selection component of the effective spread occurred, suggesting a decline in information
asymmetry among market participants. Thus, small retail investors and limit order traders are
encouraged to submit orders, thereby increasing market activity, market liquidity, and,
ultimately, market quality. We also consider the impact of the introduction of the Price
Improvement Rule on adverse selection, controlling for the possible influence of other wellknown factors (see Heflin and Shaw (2001)) and internalization (Table 6). For trades of 5000
shares or less, trading volume appear to significantly influence adverse selection in the expected
directions. The introduction of the Price Improvement Rule is associated with a significant
reduction in the adverse selection component of the effective spread, a result consistent with
reduced opportunities for informed traders to profit at the expense of uninformed traders. We
also note that for trades of 5000 shares or less internalization, by itself, does not appear to drive
adverse selection, reinforcing the notion that internalization may be associated with a high
quality market. We thus reject H3 at 5% level of significance, and conclude that the introduction
of the Price Improvement Rule significantly influences the extent of adverse selection at the
TSE.
We consider any additional insight provided on adverse selection by comparing effective
spreads with realized spreads in Table 7. The realized spread measure is similar to the effective
spread measure except that it uses the midquote 5 minutes after the trade is executed, and
captures non –information related execution costs. Thus, the difference between the effective
spread and the realized spread is considered to be a measure of the adverse selection component
of the spread. Although both realized and effective spreads decline significantly after the
introduction of the new rule, realized spreads are consistently greater than effective spreads—a
21
result consistent with findings of Jain (2004). The TSE is primarily an electronic trading venue.
Domowitz (1999) suggests that in such cases, the market structure is best analyzed by focusing
on trading products than on trader behavior. Thus, the strict interpretation of the realized spread
as a measure of adverse selection may not be particularly useful in this instance.
Internalization
Table 8 describes the effect of the rule change on internalization rates for trades in
various size ranges. Overall, small trades of 1,200 shares or less (Panels A), and trades occurring
at the spread (Panel B, except for trades between 1200 and 5,000 shares) show strong increases
in internalization rates. The increase in internalization rates for small trades suggests members’
attempt to profit by improving the efficiency of trading where possible. Since they are required
to improve on the best-quoted prices if the trade is internalized off the book, members would
tend to prefer internalization on the book, where they may still be able to internalize at the
spread. Otherwise, members would offer competitive prices on retail trades and eke out possible
profits by internalization off the book and improving on price; or internalizing off the book at the
spread if market conditions justify such an action. Consequently, the client receives a price at
least as good as the price available in a public market, unless market conditions suggest
otherwise.
Table 9 shows a strong, negative association between effective spreads and
internalization rates for trades of size 5,000 shares or less, after controlling for the influence of
price, trading volume, return volatility, and the introduction of the Price Improvement Rule on
effective spreads. This result directly contradicts the findings of Chung et al (2004), and suggests
22
that the practice of internalization can, under certain circumstances be justified by market
conditions, and may actually enhance market quality.
Market Depth
Table 10 gives the results of tests for a change in market depth following the rule change
after controlling for the influence of other well-known factors. It is expected that if market
quality improves due to the introduction of the new rule, then one unambiguous manifestation of
the resulting higher market liquidity would be lower quoted spreads accompanied by higher
market depth. Market depth is also expected to increase with trading volume and decline with
higher return volatility. The regression coefficients estimates assume the expected signs.
However, we record an insignificant association between quoted spreads and depth, a strong,
positive association between depth and trading volume, and a marginally significant, negative
relation between market depth and return volatility. Most important, a significant, positive
association between market depth and the new rule appears to exist. We thus reject H4, and
conclude that the implementation of the Price Improvement Rule significantly influences market
depth.
Results from Test for a Change in Short Term Stock Return Volatility
Table 11 provides regression results showing the effect of the rule change on short term return
volatility, controlling for the influence of trading activity. Overall, and for all volume quartiles,
strong declines in market volatility occurred, as evidenced by negative and statistically
significant dummy variable coefficient values. We reject H5, and conclude that the introduction
23
of the price improvement Rule has reduced market uncertainty by gaining the confidence of
investors in the market place, and encouraging them to offer more liquidity.
Results from a Test for a Change in the Variance of the Pricing Error
Table 12 describes the distribution of F–values for selected stocks. The F-value compares
the variance of the pricing error after the introduction of the Price Improvement Rule with the
variance before the event. A much higher proportion of stocks had F-values significantly less
than one (69.38 %) than greater than one (8.91%). More concretely, the Wilcoxon one-tailed
signed-ranks test for an overall decline in market pricing-error variance after the introduction of
the Price Improvement Rule shows a strong decline in the variance of the pricing error overall,
and results in p-values below 0.001 for the two highest volume quartiles. For the second highest
volume quartile, the p-value is 0.005. However, for low volume stocks the improvement was
marginal, with the test generating a p-value of 0.075. The results of the Wilcoxon signed-ranks
test are generally indicative of a more efficient price discovery process and a better quality
market in the post event period. The introduction of the Price Improvement Rule is associated
with a significant improvement in overall market quality at the TSE. Thus, we reject H6.
VI. Summary and Conclusions
This study assesses the outcome of attempts to improve market quality by reducing the
potentially harmful market fragmentation, resulting from the practice of internalization. The
results of an empirical investigation into the effects of the introduction of regulation by the
Toronto Stock Exchange to centralize trading indicate a generally successful effort: spreads fell,
volatility declined, and over market quality improved. Overall, the implementation of the Price
24
Improvement Rule reduced opportunities for profiting at the expense of limit order traders,
restored time priority, and increased price competition and order interaction among market and
limit orders. A particularly interesting finding is the strong, negative association between
internalization rates spreads, suggesting an effective positive association between the practice of
internalization and market quality. Thus, the disadvantages (fragmentation, loss of liquidity)
brought about by the practice of internalization appear to be outweighed by the advantages
(providing investor clienteles with ‘best’ execution most closely aligned with each clientele’s
preferences). This favorable outcome appears to result from judiciously regulated internalization
practiced under favorable market conditions. Financial markets may wish to carefully consider
the role of the practice of internalization in a given market setting before enacting sweeping
legislation designed to curb it.
25
References
Battalio, R., 1997, Third market broker-dealers: Cost competitors or cream skimmers? Journal of
Finance 52, 341-52.
Battalio, R. and R. Jennings, 1997, Do competing specialists and preferencing dealers affect
market quality? Review of Financial Studies 10, 969-93.
Battalio, R., J. Green, and R. Jennings, 1998, Order flow distribution, bid-ask spreads, and
liquidity costs: Merrill Lynch’s decision to cease routinely routing orders to regional
stock exchanges, Journal of Financial Intermediation 7, 338-58.
Battalio, R., B. Hatch, and R. Jennings, 2001, Post-reform market-order execution quality:
Multidimensional comparisons across market centers, Financial Review 36, 123-52.
Battalio, R., B. Hatch, and R. Jennings, 2003, All else equal?: A multidimensional analysis of
retail, market order execution quality, Journal of Financial Markets 6, 143-62.
Bessembinder, H., 1999, Trade execution costs on NASDAQ and the NYSE: A post-reform
comparison, Journal of Financial and Quantitative Analysis 34, 387-407.
Booth, G., J. Lin, and G. Sanger, 1995, Trade size and components of the bid-ask spread, Review
of Financial Studies 8, 1153-83.
Chakravarty, S. and A. Sarkar, 2002, A model of broker’s trading, with applications to order
flow internalization, Review of Financial Economics 11, 19-36.
Chordia, T. and A. Subrahmanyam, 1995, Market making, the tick size, and payment-for- order
flow: Theory and evidence, Journal of Business 68, 543 – 75.
Chordia, T., R. Roll, and A. Subrahmanyam, 2000, Co-movements in bid-ask spreads and market
depth, Financial Analyst Journal 56, 23 – 27.
Chung, K., C. Chuwonganant, and D. T. McCormick, 2004, Order preferencing and market
quality on NASDAQ before and after decimalization, Journal of Financial Economics
71, 581-612.
Domowitz, I. and B. Steil, 1999, Automation, trading costs, and the structure of the trading
services industry, Brookings-Wharton Papers on Financial Services.
Easley, D. and M. O’Hara, 1987, Price, trade size, and information in securities markets, Journal
of Financial Economics 19, 69-90.
Hamilton, J. L., 1979, Marketplace fragmentation, competition, and the efficiency of the stock
exchange, Journal of Finance 34, 171 – 87.
26
Hansch, O., N. Y. Naik, and S. Viswanathan, 1999, Preferencing, internalization, best execution,
and dealer profits, Journal of Finance 54, 1799-1828.
Harris, L., 1994, Minimum price variations, discrete bid/ask spreads and quotation sizes, Review
of Financial Studies 7, 149-78.
Harris, M. and A. Raviv, 1993, Differences of opinion make a horse race, Review of Financial
Studies 7, 149-178.
Hasbrouck, J., 1993, Assessing the quality of a security market: A new approach to transactioncost measurement, Review of Financial Studies 6, 191-212.
Heflin, F. and K. W. Shaw, 2001, Adverse selection, inventory-holding costs, and depth, Journal
of Financial Research 24, 65-82.
Jain, P., 2004, Institutional design and liquidity on stock exchanges around the world, Working
paper, University of Memphis.
Jones, C. M., G. Kaul, and M. L. Lipson, 1994, Transactions volume, and volatility, The Review
of Financial Studies 7, 631– 51.
Kluger, B. D. and S. B. Wyatt, 2002, Preferencing, internalization of order flow, and tacit
collusion: Evidence from experiments, Journal of Financial and Quantitative Analysis
37, 449-69.
Kumar, R., A. Sarin, and K. Shastri, 1998, The impact of options trading on the market quality of
the underlying security: An empirical analysis, Journal of Finance 53, 717-32.
Lee, C. M., B. Mucklow, and M. J. Ready, 1993, Spreads, depths, and the impact of earnings
information: An intraday analysis, The Review of Financial Studies 6, 345-74.
Madhavan, A., D. Porter, and D. Weaver, 2001, Pre-trade transparency, Istanbul Stock Exchange
Review 5, 23-46.
Peterson, M. A., and D. Fialkowski, 1994, Posted versus effective spreads, Journal of Financial
Economics 35, 269-292.
Peterson, M. A. and E. R. Sirri, 2003, Order preferencing and market quality on US Equity
Exchanges, Review of Financial Studies 16, 385-415.
Stoll, H. R., 1989, Inferring the components of the bid-ask spread: Theory and empirical
evidence, Journal of Finance 44, 115-34.
Stoll, H. R., 2001, Market fragmentation, Financial Analysts Journal 57, 16-20.
27
Toronto Stock Exchange (TSE) Special Committee Report, 1997. Market Fragmentation:
Responding to the Challenge.
28
TABLE 1. Descriptive Statistics.
Volume Quartiles
All Firms
(Lowest)
1
2
3
(Highest)
4
258
64
65
65
64
C$22.34
C$27.03
C$18.70
C$20.75
C$22.96
2,259
124
461
1,335
7,159
Minimum Volume
0
0
1
1
23
Maximum Volume
98,139
Number of Stocks
Mean Price
Mean Share Volume
Percentage of Trades
Internalized
2,307
11,156
35,510
98,139
11.18%
11.59
12.53
11.81
10.94
Total
890,404
27,896
57,103
114,965
690,440
 1,200
706,915
25,673
49,715
96,078
535,449
> 1,200, but  5,000
156,566
1,961
6,233
15,320
133,052
> 5,000
26,923
262
1,155
3,567
21,939
Number of Trades (in shares):
Note: Table 1 provides a breakdown of a sample of equity trades on the Toronto Stock Exchange that occurred prior to the
introduction of the Price Improvement Rule on October 26, 1998. The pre-event period consists of 25 trading days of trades from
September 3rd to October 8th 1998, 11 days prior to the introduction of the Price Improvement Rule. Internalized trades occur
where the same brokerage firm is on both sides of the trade. Prices are expressed in Canadian Dollars (C$). Trades eligible for
price improvement have a (quoted) tick size greater than the minimum tick size (equal C$0.05 when the stock price is greater
than C $5.00). Share volume is average daily trading volume (in hundreds of shares).
29
TABLE 2. Percentage of Trades Receiving Price Improvement.
 1,200 Shares
> 1,200,
 5,000 Shares
 5,000 Shares
18.84%
16.63%
-2.62
-2.83***
Sign Test
p-value
< 0.01
28.82
26.51
-2.31
-2.01***
< 0.01
20.58
17.89
-2.69
-3.23***
< 0.01
> 5,000 Shares
33.97
29.51
-4.46
-3.48***
< 0.01
(N = 258)
Pre
Post
Change
t-statistic
Note: This table shows the results of a test of whether a significant change in the price improvement rate occurred for various
types of trades, after the introduction of the Price Improvement Rule, using the sample described in Table 1. A price-improved
trade occurs within the inside quote. The price improvement rate for a particular type of trade is defined as the ratio of priceimproved trades of that type to all trades of that type, and is done on a stock-by-stock basis. A paired t-test is used to test for
significant changes in price improvement rates after the rule change. A nonparametric paired observations test, the Wilcoxon
Sign Test, is also conducted.
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
30
TABLE 3. Time-Weighted Quoted Spreads.
(N=258)
Price
Volume
Volatility
Dummy
F-Statistic
{adjusted R2}
Quoted Spread
($)
0.2410
-0.09372
3.2218
-0.0553
113.11
(10.63***)
(-10.45***)
(12.32***)
(-2.05**)
{0.4655)
-0.0067
-0.0034
0.0211
-0.0028
166.02
(-14.42***)
(-18.57***)
(3.92***)
(-5.03***)
{0.5617}
Quoted Spread
(%)
Note: This table shows the results of a test of whether a significant change in quoted spreads occurred for various types of trades
after the introduction of the Price Improvement Rule. The effect of the rule change on spreads is determined after controlling for
the effects of other factors. Specifically, the model is given by Lit = 0 + 1lnVit +2it2 + 3Pit + 4D_RULECHit + it , where
Lit = quoted spread measure for firm i in (pre- or post-event) period t , lnVit = the log of average daily trading volume (number of
shares) for firm i in period t , it2 = the return variance for firm i in period t , Pit = daily average price for firm i in period t ,
D_RULECHit = dummy variable with a value of 1 in the post-event period, 0 otherwise, and it = error term, assumed to have
distribution N (0, ).
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
31
TABLE 4. Share-Weighted Effective Spreads.
Price
Volume
Volatility
Dummy
F-Statistic
{adjusted R2}
0.0957
-0.0406
1.1455
-0.0201
100.75
(11.20***)
(-11.43***)
(11.37***)
(-2.61**)
{0.4290}
0.1064
-0.0430
1.1175
-0.0109
107.00
(12.07***)
(-12.12***)
(11.45***)
(-2.94***)
{ 0.44}
0.0829
-0.0256
0.3066
-0.0157
127.52
(9.62***)
(-7.85***)
(15.28***)
(-2.48**)
{0.50}
0.1080
-0.04397
1.10432
- 0.0479
104.53
(12.15***)
(-11.83***)
(11.27***)
(-3.84***)
{0.44}
0.0328
-0.0198
0.0072
-0.01540
40.93
(7.20***)
(-10.81***)
(0.43)
(-2.84**)
{0.2632}
0.0336
-0.0206
0.0108
-0.0169
37.79
(7.07***)
(-10.26***)
(0.49)
(-3.01***)
{0.25}
0.0449
-0.00570
0.4839
-0.0226
393.42
(5.74***)
(-1.78*)
(34.30***)
(-2.40**)
{0.77}
Overall (N=258)
 5,000 Shares
>5,000 Shares
 1,200 Shares
> 1,200 Shares,
5,000 shares
> 5,000 Shares,
< 10,000 Shares
 10,000 Shares
Note: This table shows the results of a test of whether a significant change in effective spreads occurred for various types of
trades after the introduction of the Price Improvement Rule using sample information described in Table 1. The effect of the rule
change on spreads is determined after controlling for the effects of other factors. Specifically, the model is given by Lit = 0 +
1lnVit +2it2 + 3Pit + 4D_RULECHit +it , where Lit = share-weighted dollar or percentage effective spread measure for firm
i in (pre- or post-event) period t , lnVit = the log of average daily trading volume (number of shares ) for firm i in period t , it2 =
the return variance for firm i in period t , Pit = daily average price for firm i in period t , D_RULECHit = dummy variable with a
value of 1 in the post-event period, 0 otherwise. Where applicable, for a given set of categories, included stocks trade shares in all
categories.
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
32
TABLE 5. Adverse Selection Component of Spread.
Pre
Post
Change
t-statistic
Overall
Volume Quartile 1
(lowest)
Volume Quartile 2
20.58
15.95
-4.63
-4.35***
32.11
28.56
-3.55
-0.41
23.33
15.22
-8.11
-3.12***
Volume Quartile 3
Volume Quartile 4
(highest)
21.44
16.86
-4.92
-2.10**
16.82
13.88
-2.94
-3.62***
Note: This table shows the results of a test of whether a significant change in the percentage adverse selection component
occurred for various trading volume groups and intraday time intervals after the introduction of the Price Improvement Rule,
using 152 stocks from the original sample (described in Table 1). Each included stock had at least a total of 50 trades in the first
and last half –hours of trading. The percentage adverse selection component of the effective spread, , is estimated by the
following regression: Qt =  z t-1+et, where Qt = Qt - Qt-1, Qt is the log mid quote at time t, z t = Pt – Qt,, and Pt is the log
transaction price at time t.
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
33
TABLE 6. Adverse Selection: Regression Results.
(N=258)
 5,000
Shares
> 5,000
Shares
Volume
Volatility
Price
Dummy
Internalization
F–Statistic
{ adjusted R2 }
-0.0257
10.3869
-4.55E-8
-0.0235
-0.0002
11.90
(-6.56***)
(2.81***)
(0.00)
(-2.06***)
(-0.30)
{0.0931)
-0.0086
5.2493
-0.0003
0.0114
0.0007
166.02
(-1.86*)
(1.25)
(-1.34)
(-0.74)
(1.79*)
{0.5617}
Note: This table shows the results of a test of whether a significant change in the adverse selection component of the spread
occurred , after the introduction of the Price Improvement Rule, controlling for other potentially influential factors. The model is
given by ASit = 0 + 1lnVit +2it2 + 3Pit + 4D_RULECHit+ IR +it , where ASit = the percentage adverse selection
component of the effective spread for firm i in (pre- or post-event) period t , lnVit = the log of average daily trading volume
(number of shares) for firm i in period t , it2 = the return variance for firm i in period t , Pit = daily average price for firm i in
period t , D_RULECHit = dummy variable with a value of 1 in the post-event period, 0 otherwise, IR = the internalization rate for a
particular type of trade is defined as the ratio of internalized trades (trades with the same buyer and seller) of that type to all
trades of that type, and it = error term, assumed to have distribution N (0, ).
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
34
TABLE 7. Effective and Realized Dollar Spreads.
Pre
Overall (N=258)
Volume Quartile 1 (lowest)
Volume Quartile 2
Volume Quartile 3
Volume Quartile 4 (highest)
Post
Change
t-statistic
Effective
C$0.1145
C$0.1014
-C$0.0131
-4.41***
Realized
C$0.1389
C$0.1186
-C$0.0203
-7.25***
Effective
C$0.1860
C$0.1674
-C$0.0186
-4.64***
Realized
C$0.2257
C$0.1886
-C$0.0371
-8.66***
Effective
C$0.0893
C$0.0852
-C$0.0041
-2.55***
Realized
C$0.1097
C$0.0993
-C$0.0104
-6.12***
Effective
C$0.0933
C$0.0770
-C$0.0163
-3.99***
Realized
C$0.1119
C$0.0955
-C$0.0164
-4.81***
Effective
C$0.0904
C$0.0765
-C$0.0139
-3.56***
Realized
C$0.1092
C$0.0918
-C$0.0174
-3.99***
Note: This table shows the results of a test of whether significant changes occur in (Canadian) dollar effective and realized
spreads after the introduction of the Price Improvement Rule. The dollar effective spread is defined as Effective Dollar Spreadit =
|Priceit – (Askit + Bidit)/2| , where Askit and Bidit are the time-weighted average ask and bid quotes for stock i in measurement
period t. and Priceit is the average transaction price of stock i in measurement period t. The realized spread measure is identical to
the effective spread measure except that it contains the midpoint of the quoted spread five minutes after the transaction is
consummated.
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
35
TABLE 8. Internalization Rates Around the Introduction of the Price Improvement Rule.
Panel A. Overall Internalization Rates
(N=258)
 1,200 Shares
> 1,200 Shares,
 5,000 shares
> 5,000 Shares
Pre
Post
Change
t-statistic
Sign Test p-value
9.85
10.89
1.04
4.34***
< 0.01
14.30
13.65
-0.65
-1.11
= 0.34
65.60
64.73
-0.87
-0.68
= 0.45
Change
t-statistic
Sign Test p-value
Panel B. Internalization Rates by Spread Location
(N=258)
 5,000
Shares
> 5,000
Shares
 1,200
Shares
> 1,200
Shares,
 5,000
shares
Spread
Location
At Spread
Inside
Spread
Outside
Spread
At Spread
Inside
Spread
Outside
Spread
At Spread
Inside
Spread
Outside
Spread
At Spread
Inside
Spread
Outside
Spread
Pre
Post
10.75
11.61
0.86
3.56***
<0.01***
10.41
10.59
0.18
0.56
=0.20
13.50
12.69
-0.81
-0.67
= 0.19
66.06
64.07
-1.99
-1.32
=0.20
64.11
64.51
3.38
0.23
=0.44
53.59
54.38
0.79
0.14
=0.49
10.00
11.31
1.31
5.66***
<0.01***
9.85
10.04
0.19
0.57
=0.19
14.11
12.31
-1.80
-1.23
= 0.31
16.86
14.58
-2.28
-3.08***
<0.01***
15.52
14.90
-0.62
-0.84
=0.45
18.73
19.85
1.12
0.30
=010
(Continued)
36
TABLE 8. Continued.
Panel C. Percentage of Internalized Trades by Spread Location
(N=258)
 5,000
Shares
> 5,000
Shares
 1,200
Shares
> 1,200
Shares,
 5,000
shares
Spread
Location
At Spread
Inside
Spread
At Spread
Inside
Spread
At Spread
Inside
Spread
At Spread
Inside
Spread
Pre
Post
Change
t-statistic
Sign Test p-value
78.46
81.90
3.44
3.56***
<0.01
20.10
17.40
-2.71
-3.02***
<0.01
66.90
71.27
4.37
3.66***
<0.01
31.64
28.20
-3.44
-3.44***
<0.01
79.68
83.46
3.78
4.24***
<0.01
18.18
15.90
-2.28
-3.51***
<0.01
70.82
74.93
4.11
2.32***
<0.01
23.39
24.13
0.74
0.49
0.39
Note: This table shows the results of a test of whether a significant change in the crossing internalization rate occurred for various
types of trades after the introduction of the Price Improvement Rule using the sample described in Table 1. An internalized trade
is one where the same buying and selling brokerage firm is on both sides of the trade. The internalization rate for a particular type
of trade is defined as the ratio of internalized trades of that type to all trades of that type, and is done on a stock-by-stock basis. A
paired t-test is used to test for significant changes in internalization rates after the rule change. A nonparametric paired
observations test, the Wilcoxon Sign Test, is also conducted.
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
37
TABLE 9. Internalization and Spreads.
Price
Volume
Volatility
Dummy
Internalized
Trades
F-Statistic
{adjusted R2}
0.1029
-0.0440
1.1192
-0.0221
-0.0003
73.28
(10.27***)
(-9.62***)
(10.87***)
(-1.90*)
(-0.87*)
{0.42}
0.1105
-0.0486
1.1327
-0.0293
-0.0039
427.41
(11.5***)
(-11.32***)
(11.15***)
(-2.53***)
(-2.99***)
{0.45}
0.0378
-0.0060
0.5310
-0.0215
0.0002
562.73
(5.01***)
(-1.17)
(41.69***)
(-1.44)
(1.02)
{0.82}
0.1096
-0.0495
1.0803
-0.0289
-0.0032
77.22
(11.67***)
(-10.83***)
(10.99***)
(-2.55***)
(-2.51***)
0.45
0.0636
-0.0282
1.0387
-0.0201
-0.0011
138.37
(7.20***)
(-8.08***)
(20.77***)
(-1.99***)
(1.54)
(0.60}
0.0381
-0.0057
0.5329
-0.0215
-0.0002
439.61
(8.35***)
(-2.01***)
(40.81***)
(-2.75***)
(0.69)
{0.82}
Overall (N= 258)
 5,000 Shares
>5,000 Shares
 1,200 Shares
> 1,200 Shares,
5,000 shares
> 5,000 Shares
Note: This table shows the results of a test of the impact of internalization on spreads after controlling for the impact of other
factors including the introduction the Price improvement Rule, using sample information described in Table 1. Specifically, the
model is given by Lit = 0 + 1lnVit +2it2 + 3Pit + 4D_RULECHit +5IRit +it , where Lit = share-weighted dollar effective
spread measure for firm i in (pre- or post-event) period t , lnVit = the log of average daily trading volume (number of shares ) for
firm i in period t , it2 = the return variance for firm i in period t , Pit = daily average price for firm i in period t , D_RULECHit =
dummy variable with a value of 1 in the post-event period, 0 otherwise, and IRit = fraction of all trades that are internalized for firm i
in period t .Where applicable, for a given set of categories, included stocks trade shares in all categories.
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
38
TABLE 10. Time-Weighted Depth Regression Analysis.
Spread
-3.99
(-1.39)
Volatility
9.48
-3038.46
4.54
52.94
(-1.95*)
(1.99**)
{0.28}
(12.67***)
Dummy
F-Statistic
{adjusted R2}
Volume
Note: This table shows the results of tests of the effect of the introduction of the Price Improvement Rule on market depth using sample
information described in Table 1. The strength of this effect is determined after controlling for the effects of other factors. Specifically, the model
is given by Depthit = 0 +1Lit+2lnVit +3it2 + 4D_RULECHit +it , where Depthit is the time-weighted average of the arithmetic average of
the number of shares offered at the bid price and the number of shares offered at the ask price for firm i in period t (pre- or post-event period), Lit
is the time-weighted daily average quoted spread for firm i in period t, lnVit = the log of average daily trading volume (number of shares ) for firm i in
period t , it2 = the return variance for firm i in period t , and D_RULECHit= dummy variable with a value of 1 if the observation is in the post-event
period, and 0 otherwise; and it = error term, assumed to have distribution N (0, ).
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
39
TABLE 11. Volatility.
Overall
Volume Quartile 1
(lowest)
Volume Quartile 2
Volume Quartile 3
Volume Quartile 4
(highest)
Average Daily Number of
Trades
Dummy
F–Statistic
{ Adjusted R2 }
-2.16 E-6
-1.58 E-3
61.62
(-9.96***)
(-4.98***)
{0.19}
-2.1 E-6
-1.98 E-3
7.27
(-2.22**)
(-2.75***)
{0.10}
-2.96 E-6
-1.15 E-3
29.33
(-6.99*)
(-1.90*)
{0.32}
-1.03 E-6
-1.09 E-3
39.78
(-8.20***)
(-2.68***)
{0.39}
-7.85 E-8
-7.39 E-4
32.40
(-7.68***)
(-2.92***)
{0.34}
Note: This table shows the results of a test of whether a significant change in return volatility occurred for stocks in various
volume groups (described in Table 1) after controlling for the influence of trading activity. Specifically, the following regression
model is run:it = 0 +1N_Tradesit+2D_RULECHit +it, where it is the standard deviation of returns for firm i in
measurement period t (pre- or post event period), N_Tradesit is the number of transactions for firm i during measurement period
t, and D_RULECHit is a dummy variable assigned the value of 1 if the observation is from the post-event period, 0 otherwise,
and it is an error term.
***Significant at the 1% level.
**Significant at the 5% level.
*Significant at the 10% level.
40
TABLE 12. Impact of Rule Change on the Variance of the Pricing Error.
Total number of stocks
258
Mean VR-value
0.99
Median VR-value
0.67
Percentage of stocks with VR-value significantly (at 5% level) less than one
69.38
Percentage of stocks with VR-value significantly (at 5% level) greater than one
8.91
Percentage of statistically insignificant (at 5% level) VR-values
21.71
One-tailed signed-ranks probability for change in pricing error variance over event:
Overall
< 0.001
Volume Quartile 1 (lowest)
= 0.075
Volume Quartile 2
= 0.005
Volume Quartile 3
< 0.001
Volume Quartile 4 (highest)
< 0.001
Note: This table describes statistical results concerning the variance ratio VRi = S,A2 /S,B2, where S,B2 is the variance of the
pricing error for stock i in the pre-event period, and S,A2 is the variance of the pricing error for stock i in the post-event period.
The variance of the pricing error is estimated using Vector Autoregression (VAR) model developed by Hasbrouck (1993), and is
used as an overall measure of market quality. In particular, the standard deviation of the pricing error is used to measure how
closely the transaction price tracks the efficient price. The Wilcoxon one-tailed signed-ranks test is used to determine whether, on
average, VR is significantly less than one.