darktradingonpublicexchanges · opponents argue that dark trading removes liquidity from visible...
TRANSCRIPT
Dark Trading on Public Exchanges∗
Sean Foley†
University of SydneyKatya Malinova‡
University of TorontoAndreas Park§
University of Toronto
This version: March 15, 2013
Abstract
Over the last decade, market participants increasingly use trading tools that allowthem to hide their trading intentions. We study how “dark trading” in the formof fully hidden, or dark, orders posted on a visible exchange affects the quality ofthe visible market. Dark orders were introduced on the Toronto Stock Exchangein 2011 in two stages, allowing us to employ a difference-in-differences approach toisolate the causal effect of the availability of dark trading. Using order level data, weobserve that the introduction of dark orders led to a widening of displayed quotedspreads and an increase in trading costs, while leaving depth, overall volume, andvolatility unaffected. At the intra-day level, dark trading leads to decreased displayedquoted spreads, increased depth, increased volume and to reduced trading costs andvolatility. These findings may be interpreted as two sides of the same coin: thepossible presence of dark liquidity causes market participants to post visible quotesmore carefully. Upon detecting dark executions, however, traders infer that darkliquidity is diminished and thus may post quotes more aggressively.
∗The Toronto Stock Exchange (TSX) and TriAct Canada kindly provided us with databases. The viewsexpressed here are those of the authors and do not necessarily represent the views of the TMX Group orTriAct Canada. TSX Inc. holds copyright to its data, all rights reserved. It is not to be reproduced orredistributed. TSX Inc. disclaims all representations and warranties with respect to this information, andshall not be liable to any person for any use of this information. Financial support from the SSHRC (grantnumber 410101750) is gratefully acknowledged. We thank Gustavo Bobonis, Carole Comerton-Forde, MaoYe, and Vincent van Kervel for valuable feedback.
†[email protected]‡[email protected]§[email protected] (corresponding author)
Over the last decade an increasing fraction of equity market trading volume changed
hands without pre-trade transparency, or “in the dark”. This dark trading can occur on
separate, dedicated venues, such as so-called dark-pools, or it can occur on public venues
if these allow orders that are not visible to the market. The shift of trading to dark
has attracted the attention of regulators and policy-makers worldwide. This interest has
focussed on the implications of dark trading for market quality.1 Proponents of dark trading
argue, for instance, that the ability to use a dark trade gives larger orders the opportunity to
avoid being “front run” by sophisticated computer algorithms that can detect large orders.
Opponents argue that dark trading removes liquidity from visible (or “lit”) markets and
harms market participants by hampering the price discovery process.
The impact of dark trading remains an open question. First, little data on dark trading
has been made available. Second, new dark trading tools have little impact until sufficiently
many market participants have access to them, thus hampering event studies. Third, dark
trading may be endogenous to market conditions, as highlighted in Buti, Rindi, and Werner
(2011b), and so it is challenging to establish a causal relation without sufficiently granular
data. We now present a study that at least partially overcomes these challenges.
In this paper we analyze the introduction and intra-day usage of fully hidden orders on
the Toronto Stock Exchange (TSX) in 2011 using a proprietary order-level dataset. Dark
orders on the TSX were introduced in two steps. As of April 01, 2011, the constituents
of the TSX60 (the main index) could be traded with dark orders; starting May 20, 2011,
all remaining symbols were enabled for dark trading. The TSX is the main exchange in
Canada with more than 60% market share in trading volume, most brokers are connected
to the TSX, and dark orders can interact with non-dark orders. The staggered introduction
together with the immediate relevance of dark trading for all market participants allows us
to perform an event study, using a difference-in-differences approach to control for market-
wide fluctuations. The order-level data allows us to establish new stylized facts on the
intra-day usage of dark trading and to study the intra-day relation between dark trading
1For instance, Canadian regulators recently decided to take steps to significantly curtail dark trading(starting October 12, 2012).
1
and market quality.
Fully hidden dark orders on the TSX come in two forms. The first is a priced but
fully hidden limit order. The second, and most frequently used type (in our sample) is a
midpoint-pegged order that is priced at the (fluctuating) midpoint of the national best bid
and ask prices.2
Our main result is that the introduction of dark, on-exchange trading leads to a widen-
ing of displayed quoted bid-ask spreads and to an increase in trading costs, measured by
effective spreads, both accounting for and ignoring the trading fees that the exchange levies.
Displayed market depth, the visible amount available for trading at the most favourable
quoted prices, volume, and volatility remain unaffected. Our finding on bid-ask spreads is
in line with Buti and Rindi (2011)’s theoretical predictions.
Although the initial usage of dark order was not substantial, the bid-ask spread did
increase, which is consistent with people believing that hidden activities may occur. In
other words, our results on the introduction highlight that the possibility of trading with
dark orders may have an impact.
To observe the effects of the actual hidden activities, one needs a sufficient usage of
dark orders. To establish these effects, we perform an analysis for a later time when dark
orders were more frequently used, and we thus focus on the second half of 2011. For
this time span, dark orders could be used for all symbols. We base the intra-day part of
our analysis on the constituents of the TSX Composite (the broadest Canadian market
index) and begin by identifying a number of stylized facts. The quoted spreads that prevail
at transactions are wider when the liquidity providing order is a dark order than when
the liquidity providing order is a visible order; the corresponding prevailing (log-)depth
is larger with dark orders. Trading costs, both accounting for and omitting the trading
fees that the exchange levies (also referred to as maker-taker fees) are lower when dark
orders are on the liquidity providing side. This result is intuitive because trades against
2There are also so-called iceberg orders which display a portion of the total order size; these typesof orders are not the focus of our study as their functionality is fundamentally different from that offully hidden orders. The latter may execute ahead of all visible orders due to price priority, whereas theundisplayed portion of an iceberg order always executes after the visible portion.
2
dark orders may occur “inside” the bid-ask spread. The information content of trade
executions against liquidity-providing dark orders, measured by their price impact, is lower.
The fill rate, measured as the ratio of traded to submitted volume, of posted dark orders
compared to posted visible orders is higher. There are two possible explanations: first,
(midpoint) dark orders offer better prices than visible orders and thus trade ahead of these;
second, midpoint-pegged orders move with prices and thus require fewer modifications
and/or cancellations. The size of dark orders is around 2.5 to 3 times larger than visible
orders.
Next, we investigate the effect of dark trading at the intra-day level. As Buti, Rindi,
and Werner (2011b) discuss, establishing a causal relationship from dark trading to market
quality is difficult, because of an inherent endogeneity, i.e. bid-ask spreads may affect dark
trading and vice versa. We exploit our order-level data to get a step closer to establishing
the causal component of the relationship. To control for the possible interdependence of
the market quality measures (such as bid-ask spreads) and dark trading, we use the share of
dark orders one period lagged as an instrument for dark trading. It is a suitable instrument
for two reasons. First, dark orders submissions occur before dark trades. Second, dark
orders are invisible and thus have no direct effect on displayed bid-ask spreads or other
observable variables by themselves, other than through subsequent trades that lead to their
execution. Third, future spreads should not affect current order submissions.3 We find that
a larger proportion of dark trading leads to lower quoted spreads, effective spreads (with
and without taker fees), and volatility, and that it increases depth, volume, and fill-rates.
The intra-day findings are consistent with the causal effect that we attribute to the
introduction of dark orders. First, our results from the introduction of dark orders indicate
that traders post quotes more cautiously as they account for the possibility of undisplayed
liquidity. Consistent with this finding we observe the stylized fact in our data that the
prevailing displayed spread is wider at the time of a dark execution. Consequently, prior
to dark order executions, quotes were posted “cautiously”. There are a number of possible
3We acknowledge, however, that this point is debatable as spreads are autocorrelated.
3
explanations for the observed tightening of displayed spreads subsequent to dark executions.
One explanation is that upon observing executions that can be attributed to dark orders
(e.g., trades at the midpoint), traders believe that dark liquidity is diminished and are
willing to quote more aggressively. An alternative explanation is that dark liquidity, which
provides price improvement and leads to lower trading fees, is considered to be cheap so
that, upon observing a dark execution traders submit orders aggressively to consume dark
liquidity. Both explanations are consistent with our findings.
The focus of our study is dark trading on a visible exchange, but we believe that our
results provide insights on dark trading in general. More specifically, we believe that our
analysis applies to the situation where a dark pool that matches orders continuously is
introduced alongside a visible market. Assuming that trades in this new dark pool occur at
a price within the visible spread, it is cheaper to trade on the dark venue; ceteris paribus,
market-order traders should send their order to the dark venue first and then route any
unfilled portion to the visible venue.4 Liquidity providers are then in a similar situation as
in our study, since they must account for the possibility of dark liquidity that will execute
before their quotes are hit. The observations from our analysis are thus instructive to
assess the impact of such dark pool trading. Furthermore, studying trading in Canada is
instructive for the world’s largest market, the U.S., since many market players such as high
frequency trading firms are active in both markets and key features of market regulation,
such as so-called best execution and order protection, apply in both markets.
Our paper proceeds as follows: The next section reviews the literature on dark trading.
Section II describes the dark trading options that are available in Canada. Section III
describes our data, our sample, and our methodology. Section IV reviews the market
quality measures that we consider. Section V presents the results from our event study
analysis; Section VI presents the results from our intra-day analysis. Section VII concludes.
4Some dark trading venues, such as MatchNow in Canada, offer this functionality, i.e., they allow marketorders to pass through their system. A trade occurs if there is a sitting order on the other side; if not, theorder proceeds to a lit market place.
4
I Related Literature
A number of different strands of the literature address trading with limited transparency.
Iceberg Orders. One strand of the literature analyzes the impact of so-called iceberg
or reserve orders. These orders visibly display a portion of their volume. The undisplayed
portion usually has lower time priority than any visible order, and thus the visible portion
executes before the invisible one; in contrast, the midpoint-pegged orders that we study
here execute before visible orders in their entirety because of price priority. Due to this
structural difference, iceberg orders should have a different effect on visible quotes. In
this literature, Anand and Weaver (2004) study the abolishment and re-introduction of
iceberg orders on the Toronto Stock Exchange, and Aitken, Berkman, and Mak (2001),
De Winne and D’hondt (2007), Pardo and Pascual (2011), Bessembinder, Panayides, and
Venkataraman (2009) and Frey and Sandas (2009) study the usage and detectability of these
orders on the Australian Stock Exchange, Paris Euronext, Spanish Stock Exchange, Paris
Bourse, and Deutsche Boerse, respectively. Yao (2012) studies on-exchange dark trading
for NASDAQ; dark trading in her data comprises both fully hidden orders and hidden
portions of iceberg orders. She finds that executions with hidden orders yield higher profits
compared to executions with lit orders.
Dark Pools. A second strand of the literature analyzes the trading in dark pools vis-
a-vis lit markets. Studying monthly trading data for NASDAQ stocks from the buy-side
dark pools Liquidnet and ITG’s POSIT from June 2005-September 2007, Ready (2009)
finds usage of these dark pools is positively related to dollar spreads, but that usage for
stocks with higher percentage spreads is lower. Buti, Rindi, and Werner (2011b) use the
voluntarily reported daily Securities Industry and Financial Market Association U.S. data
and find that higher levels of trading in dark pools are associated with lower spreads, higher
depth, and lower volatility. For the Dutch market, Degryse, De Jong, and Van Kervel
(2011) find that the fraction of dark pool trades has the opposite effect on market quality:
they find that a larger fraction of dark trading causes an increase in price impact and wider
quoted and effective spreads. Weaver (2011) uses U.S. data from Trade Reporting Facilities
5
(TRF) and focusses on the impact of internalized trades. He finds that a higher fraction of
TRF trades is associated with larger effective and quoted spreads and larger price impact.
Boni, Brown, and Leach (2012) study the role of exclusivity in access to dark pools and
they show that exclusivity affects execution quality. Using a proprietary dataset from an
(unspecified) U.S. crossing network, Nimalendran and Ray (2012) examine the information
flows between dark and visible venues. They show that quoted spreads and price impact
measures on visible exchanges are higher following dark executions, with the largest effects
stemming from algorithmic trades in less liquid stocks. Using data from the Australian
Stock Exchange (ASX), Comerton-Forde and Putnins (2012) examine the impact of dark
trading on price discovery. They find that the orders that are executed in the dark are
less informed and that price discovery deteriorates when dark volume exceeds 10%. Dark
trading in Comerton-Forde and Putnins (2012) includes broker crosses (which we exclude),
the ASX’s dark pool (introduced in 2010), and broker-operated dark pools.
Theory. There are a number of theoretical contributions that provide guidance and
predictions for the impact of dark trading. The work most closely related to ours is Buti
and Rindi (2011).5 They predict that the introduction of hidden, midpoint pegged orders,
will cause spreads to widen, depth to increase, and volume to decrease. They further show
that traders use midpoint orders more intensively when volatility increases. We confirm
their hypothesis that the spread widens but find no changes for volume or depth.
Moinas (2010) studies iceberg order submission by an informed trader, and finds that
pre-trade opacity improves market quality. Boulatov and George (2013) compare two mar-
kets: in one, all liquidity is displayed, in the other all liquidity is hidden. Informed traders
compete more aggressively for liquidity provision in the all-dark market, so that the bid-ask
spread and the price impact are lower compared to the all-lit market. Using an experimental
market framework, Bloomfield, Saar, and O’Hara (2012) show that most market outcomes
are largely unaffected by the availability of hidden orders. While quoted spreads almost
double, the “true” spread (which includes dark liquidity) remains unaffected. We confirm
5The results that we are referring to are in the working paper version of their paper which discussesboth fully hidden and Iceberg orders; their published paper focusses on Iceberg orders only.
6
Bloomfield, Saar, and O’Hara (2012)’s finding that the quoted spread increases. We further
observe that effective spreads, which reflect the “true” spread, increase. The difference may
in part be due to the limited uptake of dark orders on the TSX — Boulatov and George
(2013) assume, for their all-dark market, that all liquidity provision involves hidden orders,
and Bloomfield, Saar, and O’Hara (2012) document extensive usage of hidden orders.
Ye (2011) studies dark pool trading using a Kyle framework. He identifies a positive
relation between fundamental value uncertainty and (a) the on-exchange price impact and
(b) the share of dark trading, and he identifies a negative relation between fundamental
value uncertainty and the dark fill-rate. In Zhu (2012), the choice between dark vs. lit
venue is driven by the difference in execution probabilities. Specifically, since informed
traders’ information is correlated, they trade on the same side of the market, making
the certainty of execution from submitting market orders to the limit order book more
attractive. As informed traders use the visible market, quoted spreads and price impacts
increase. Consistent with Zhu (2012) we find that the introduction of a dark venue increases
quoted spreads. Buti, Rindi, and Werner (2011a) study a situation in which a dark pool is
introduced alongside a public (limit order book) market. In their model traders can choose
only one or the other, and thus their theoretical findings do not apply to our framework.
II Dark Trading in Canada
Dark trading in Canada refers to the lack of pre-trade transparency in the sense that a
trader does not publicly reveal his or her trading intentions. Canadian regulation mandates
immediate post-trade transparency of prices and quantities for all transactions, whether
dark or lit.
Trading away from visible markets can occur in the form of broker crosses, electronic
(buy-side) crossing networks (e.g., Liquidnet), and continuous dark pools (e.g., ITG’s
MatchNow). Trading in Canada follows rules that, historically, have been designed to fos-
ter transparency and trading on marketplaces with pre-trade transparency. For instance,
7
trading in the off-exchange market, colloquially also referred to as “upstairs” trades, have
to follow a set of rules. Such orders fall under the “order exposure rule” which requires
that such orders be exposed to the public market,6 Once a cross has been arranged, it must
be reported to a public exchange immediately; Canada has no trade reporting facilities
as in the U.S. or Europe and broker crosses are usually posted on the Toronto Stock Ex-
change. Compared to the U.S. where a large fraction of retail orders is either internalized
by broker-dealers or sold in payment-for-order-flow arrangements, in Canada payment-for-
order-flow is illegal (including using U.S. based systems), and off-exchange internalization
is thus discouraged.7 To date, there are only two continuous dark pools in Canada, ITG’s
MatchNow (since 2007) and Alpha’s IntraSpread (since July 2011).
The focus of this study are undisplayed, on-exchange orders, which we refer to as “dark
orders”. Orders that have at least some displayed volume are commonly referred to as
“Iceberg” orders and they have been studied extensively in the literature. In Canada,
the Toronto Stock Exchange (TSX) and also Chi-X Canada offer the submission of priced
orders that do not display any volume. These dark orders are part of the limit order book
and they interact with incoming marketable orders. In our sample, around 65% of all dark
order volume is midpoint pegged, and 66% of executions with dark orders involve midpoint-
pegged dark orders. Midpoint orders execute at the midpoint of the national best bid and
offer price, provided that the bid-ask spread is positive. Submitters of midpoint orders
can further specify a limit price such that the order does not execute at prices outside
of that limit. Midpoint orders will generally have price priority. For same prices, visible
orders have priority over dark orders.
6An exception arises if the broker who facilitates an off-exchange trade provides price improvement.If the order is sufficiently large (more than 50 standard trading units (usually 100 shares) or $100,000 invalue), price improvement is not required, but order protection regulation still applies so that same-pricedorders on visible venues must be cleared; further details are outline in Section 6.4 of the Universal MarketIntegrity Rules (UMIR). See also Smith, Turnbull, and White (2001) for more details on the Canadianupstairs market, or Bessembinder and Venkataraman (2004) for the French upstairs market.
7For the TSX Composite stocks there are, on average, less than two dealer crosses per day with anaverage size in excess of 100,000 shares; see Table I for details.
8
III Data, Sample Selection, and Methodology
A Data
Our analysis is based on a proprietary order-level dataset, provided to us by the Toronto
Stock Exchange (TSX).8 Most visible market trading in Canada occurs on the TSX, TMX
Select, Pure, Alpha, and Chi-X; to compute market shares of these venues we obtained
volume data from SIRCA’s ThomsonReuters Tick History for these venues.9 Data on
shares outstanding, splits, and index status is obtained from the monthly TSX e-Review
publications. Data on the Canadian volatility index VIXC is from the Montreal Exchange’s
(the Canadian derivatives exchange) website. We exclude trading days that have no or
limited U.S. trading (an example is U.S. Thanksgiving and the Black Friday following it)
as there is little trading in Canada on these days. Information on scheduled U.S. market
closures is obtained from the NYSE Calendar.
Our data identifies dealer crosses, and we exclude these trades from our sample as the
timing and size of crosses is likely determined by a number of unobserved and uncontrol-
lable factors. The TSX introduced dark orders in three phases. As of March 15, 2011, three
symbols (BBD.B (Bombardier Class B shares), COS (Canadian Oil Sands), and SC (Shop-
per’s Drugmart)) could be traded with dark orders; as of April 01, 2011, the constituents of
the TSX60, Canada’s blue-chip index, and all Exchange Traded Funds (ETFs) were added;
and as of May 23, 2011, all symbols could be traded with dark orders. Our analysis then
proceeds in two steps. First, to understand the impact of the introduction, we perform an
event study for the April 01 adoption of dark orders for TSX60 symbols, using an event
window from January 01, 2011 to May 20, 2011. Second, to complete our understanding
of the effect of the actual usage of dark orders we study the intra-day occurrence of dark
trading. For this analysis to be meaningful we need a sufficient uptake of dark orders and
8Legal disclaimer: TSX Inc. holds copyright to its data, all rights reserved. It is not to be reproducedor redistributed. TSX Inc. disclaims all representations and warranties with respect to this information,and shall not be liable to any person for any use of this information.
9The only visible market place that our data does not cover is Omega ATS; according to the annualIIROC trading statistics (available on their website, www.iiroc.ca), Omega’s market share in 2011was below2.2% of all dollar-volume.
9
we thus perform the intra-day study for the second half of 2011 when dark trading was
available for all securities.10 For this intra-day analysis, we use only trading data between
9:45 a.m. and 3:35 p.m. to avoid potentially contaminating effects from the market opening
and the market-on-close (beginning at 3:40 p.m.).
The TSX data is the input-output of the central trading engine, and it includes all
messages from the (automated) message protocol between the brokers and the exchange.
Messages include all orders, cancellations and modifications, all trade reports, and all details
on dealer (upstairs) crosses. Each message consists of up to 500 subentries, such as the
date, ticker symbol, time stamp, price, and volume. The data identifies orders with hidden
portions (i.e., iceberg orders), dark midpoint orders, and fully hidden orders. Our focus is
on dark trading via the limit order book. Consequently, we exclude opening trades, oddlot
trades,11 dealer crosses, trades in the special terms market, and trades that occur outside
normal trading hours. The data also specifies the active (liquidity demanding) and passive
(liquidity supplying) party, thus identifying each trade as buyer- or seller-initiated. One
useful system message is the “prevailing quote”. It identifies the best bid and ask quotes
as well as the depth at the best quotes and is updated each time there is a change in the
price or depth of these quotes. Starting with the introduction of dark orders, the data
additionally allows us to determine the prevailing national best bid and offer prices.
B Sample Selection
For our analysis, we focus on individual company stocks and discard ETFs. We select
symbols from the TSX Composite index, split into a treatment group of TSX60 symbols,
and a control group, comprising the remaining symbols. This control group of companies
that are in the TSX Composite but not in the TSX60 is also referred to as the “TSX
Completion” index. We require that the companies were part of the TSX Composite for
10It is our impression that a number of traders only started using these orders after they were availablefor all securities.
11Oddlot trades are portions of orders that are not in multiples of 100 shares; these are not cleared viathe limit order book, but they are automatically cleared via the so-called registered trader for that symbol.
10
the entire sample period. We exclude securities that had stock splits (January - May for the
introduction, January-December for the intra-day analysis), that had days with an average
midprice below $1, or that had less than 10 transactions per day. We also eliminated
Fairfax Financial Holdings (FFH) which has an average share price above $400 (the second
highest priced stock in our sample has an average price of around $120). Finally, because
of damaged data files and suspected data errors, we had to eliminate January 03, October
3rd and November 30th, and, for symbol ABX only, November 09. This leaves for the
introduction/intra-day analysis, respectively, 230/229 companies; 52/53 are constituents of
the TSX60, the remainder are constituents of the TSX Completion.
C Regression Methodology for the Introduction of Dark Orders
To understand the impact of the introduction of dark orders, we perform a difference
in differences event study on the panel of TSX Composite constituents, using the TSX
Completion constituents as a control group for the TSX60 constituents to account for
market-wide fluctuations. There are caveats: TSX60 constituents are larger and trading
in these stocks is more competitive; for instance, bid-ask spreads are smaller and depth is
larger. We control for these issues by including symbol fixed effects.12
We further control for daily fluctuations in market-wide volatility by using the Canadian
volatility index VIXC. This index is based on the implied volatility of TSX60 index options.
To account for the possibility that the TSX60 and the TSX Completion constituents react
differently to changes in this index, we further estimated specifications in which we allow
for a differentiated response to the VIXCt by interacting it with a dummy for the TSX60
and for the TSX Completion.
Degryse, De Jong, and Van Kervel (2011) and O’Hara and Ye (2011) show that mar-
12Alternatively, we also ran regression specifications in which we controlled for size and a number ofother long-term, pre-event liquidity-determining measures; the results from this analysis are similar. Anumber of securities are also cross-listed with U.S. exchanges. The results for the group of cross-listedsecurities is similar to our main analysis, though the estimates have lower significance, and the resultsfor the non-cross-listed securities are inconclusive. However, there are few non-crosslisted securities in theTSX60; likewise, there are few cross-listed securities in the control group of TSX Completion securities,causing our tests to have low power. We thus omit this analysis.
11
ket fragmentation plays a major role in understanding trading costs. We follow Degryse,
De Jong, and Van Kervel (2011) and measure fragmentation as one minus the per firm
daily Herfindahl-Hirschman Index (HHI) based on the squared shares of total volume that
are traded on each venue. We use the contemporaneous observation as a security-level con-
trol variable. For our sample of TSX Composite firms in the second half of 2011, market
fragmentation was about 0.5; see Table I.
To avoid biased standard errors stemming from observations being correlated across
time by firm or across firms by time or both, we follow the procedure suggested in Petersen
(2009) and compare a variety of specifications to determine which fixed effects to include
and on which variables to cluster the standard errors. Based on this approach, we employ
standard errors that are clustered by both firm and date and firm-fixed effects for all
specifications in this paper.13
Part 1: Event Study. In the first part of our analysis, we perform an event study on
the effect of the introduction of dark orders. We base our analysis on the daily averages of
a number of key measures. For each measure, we run the following regression
dependent variableit = αi+ β1× eventt × tsx60i + β2× eventt+
n∑
j=1
β2+jcontrolijt+ ǫit, (1)
where dependent variableit is the day t realization of the measure for security i; αi are
firm fixed effects; eventt is an indicator variable that is 1 after the event date, April 1st,
2011, and 0 before; tsx60i is an indicator variable that is 1 for constituents of the TSX60
(the treatment group) and 0 otherwise; and controlijt are daily realizations of security
level control variables for the company, namely, depending on the specifications, the log
of the average daily mid-quote, the level of market fragmentation, and the Canadian VIX
(possibly interacted with the index constituency indicator). The variable of interest in our
regressions is the coefficient on the event-treatment group interaction term, β1.
13Cameron, Gelbach, and Miller (2011) and Thompson (2011) developed the double-clustering approachsimultaneously. Controlling for autocorrelation of the errors for up to 5 lags had no impact on the resultsand we thus present the results without this correction.
12
To further understand the effect of a level-change in dark trading on the measures
of interest, we analyze a specification in which we include the daily percentage of dark
volume, defined as the percent of daily TSX trading volume for which at least one side
of a trade is a dark order. The specification is estimated using a 2-stage approach, where
we instrument the percent of dark trading by the interaction of the event with index
constituency, eventt × tsx60i:
dependent variableit = αi + β1 ×% dark tradingit + β2 × eventt
+β3VIXCt +∑n
j=1 β3+jcontrolijt + ǫit.
The estimated coefficient then is interpreted as signifying the linear effect on the dependent
variable caused by a 1% increase in dark trading. We ran a number of tests to ensure
that the specification is not misspecified and compute the Kleibergen and Paap (2006)
Wald statistic of under-identification, which, for our specification, is χ2(1) distributed, the
Kleibergen and Paap (2006) statistic for weak identification (critical values are by Andrews
and Stock (2005)); since we use a single instrument only it is not necessary to test for over-
identification.
In principle, the introduction of dark trading for non-TSX60 stocks after May 20, 2011,
lends itself to studying this second event. However, a simple event study with a regressor
such as second event 2 × non-TSX60 stock is difficult to interpret. First, the adoption of
dark trading took time, and so the extent of dark trading in the TSX60 stocks continued to
increase past the second event data, that is, after May 20, dark trading changed for both
groups. Second, dark trading for TSX60 stocks may also be affected directly by the second
event. An example is a situation where a group of traders trades in TSX60 and non-TSX60
stocks but waits until the second event before starting to use dark trading, e.g. because of
technological limitations. With these caveats in mind, we estimated a specification similar
to (2) where we instrument the percent of dark trading by the interaction of the two events
with the affected index constituents, event1t × tsx60i and event2t ×non-tsx60i. However, for
this specification, many estimations led to a rejection of the over-identification test, based
13
on Hansen’s J-stat, indicating that the IV coefficients did not vary significantly when using
different subsets of the excluded IV sets. This finding confirmed our economic intuition
that there was a spill-over effect to TSX60 stocks following the second event, making it
impossible to interpret the IV coefficient as the effect of the second event on the outcomes
for non-TSX60 constituents. We thus do not present the results for these estimations.
Part 2: Intra-Day Analysis. To further our understanding of the impact of dark
trading, we seek to determine how the occurrence and extent of dark trading relates to and
affects liquidity. We are particularly interested in assessing whether dark trading has an
impact on contemporaneous liquidity.
As Buti, Rindi, and Werner (2011b) argue, there may an endogenous relationship be-
tween market quality variables and dark trading activity. To get a step closer to making
a causal inference from dark trading to market quality, we must identify a plausible in-
strument for dark trading, that is, a variable that is highly correlated with dark trading
but that is not affected by the contemporaneous spread. We believe that the fraction of
dark order volume at time t− 1 is a reasonable instrument for the fraction of dark trading
volume at time t, with t being measured intra-day, where we define dark trading volume
as the volume of trades for which a dark order was on the passive side. The intuition is as
follows: lagged dark orders correlate with dark trading because orders precede executions.
Since dark orders are invisible until they are part of a trade, they cannot directly affect
market quality measures that are based on visible measures. Consequently, dark order vol-
ume should affect market quality variables only through dark trading. Finally, dark order
volume at time t− 1 is, arguably, not affected by future bid-ask spreads. While we believe
that our approach is valid, there is a caveat: liquidity variables such as the bid-ask spread
are usually autocorrelated throughout the day and so when submitting a dark order at
time t− 1, traders may form expectations about the spreads at t.
As in the first part of our study, we use firm fixed effects and we double-cluster the
standard errors by time and firm. Figures 4 and 5 show that there are intra-day patterns
14
in spreads, depth, volume, and volatility. To control for the possibility that the intra-
day patterns of observable variables affect outcomes, we include time-of-day fixed effects.
We analyze the impact in a number of specifications, using different controls such as the
log-midquote, market fragmentation, and the daily realization of the Canadian VIX. Fur-
thermore, Degryse, De Jong, and Van Kervel (2011) suggest a number of contemporaneous
stock-level control variables, such as volatility (we use the Hasbrouck and Saar (2012) mea-
sure of maximum relative price fluctuation in a 10-minute time interval), volume, fill-rate
(trade volume relative to order volume), and the average execution size (measured over
a long horizon). Finally, we use the average of volatility, volume, and trade-size over all
stocks except for i as firm-level controls.
We split the trading day into 35 10-minute intervals between 9:45 a.m. and 3:35 p.m. and
compute average liquidity measures and aggregate volumes over these intervals. Shorter
intervals may result in many empty observations for dark trading, whilst longer time in-
tervals eliminate the justification for the relationship between dark orders and dark trades.
We then analyzed the following specification using two-stage least square regressions
dependent variableit = αi+γ(k)+β1×% dark tradingit+β2VIXCd+n∑
j=1
β3+jcontrolijt+ǫit (2)
where αi and γ(k) are stock and 34 time-of-day fixed effects respectively,14 VIXCd is the
daily realization of the Canadian VIX, and controls are as discussed above; the fraction of
trading involving dark orders in stock i for time interval t, % dark tradingit, is instrumented
by % dark orderit−1, the fraction of dark orders, one period lagged, in the first stage of the
regression. Since, economically, % dark orderit−1 is not defined for the first interval of the
day, we drop the first observation for each day. We further test for misspecification in the
first step of 2-stage regressions and compute the Kleibergen and Paap (2006) Wald statistic
of under-identification and the Kleibergen and Paap (2006) statistic for weak identification;
since we use a single instrument, a test for over-identification is unnecessary.
14In untabulated regressions we also included day-dummies in addition to the other dummies; the resultsare similar to the ones that we present both in terms of magnitude of the effects and in terms of statisticalsignificance.
15
IV Liquidity Measures
A Quoted Visible Liquidity
We measure quoted liquidity using displayed time weighted quoted spreads and depth. The
quoted spread is the difference between the lowest price at which someone is willing to sell,
or the best offer price, and the highest price at which someone is willing to buy, or the best
bid price. We focus on the spread measures expressed in cents. Share depth is defined as
the average of the number of shares that can be traded on the bid and offer side; the dollar
depth is the value that can be traded at the bid and the offer. We use logarithms of the
depth measures to ensure a more symmetric distribution since several Canadian companies
historically have very large depth. High liquidity refers to large depth and small spreads.
The time weighted measures reflect the availability of liquidity throughout the day.
For our event study approach, we focus on the effect on the TSX spread, i.e., the best
bid-ask spread quoted on the TSX alone. For our intra-day analysis, we compute both the
TSX-only and the Canada-wide spread. We compute depth only for the TSX as we do not
have depth data for the other trading venues.
B Effective Liquidity with and without Maker-Taker Fees
Quoted liquidity only measures posted conditions, whereas effective liquidity captures the
conditions that traders decided to act upon. The costs of a transaction to the liquidity
demander are measured by the effective spread, which is the difference between the transac-
tion price and the midpoint of the bid and ask quotes at the time of the transaction. This
measure also captures the costs that arise when the volume of an incoming order exceeds
the posted size at the best prices. For the t-th trade in stock i, the effective spread is
defined as
espreadti = 2qti(pti −mti), (3)
16
where pti is the transaction price, mti is the midpoint of the quote prevailing at the time
of the trade, and qti is an indicator variable, which equals 1 if the trade is buyer-initiated
and −1 if the trade is seller-initiated. Our data includes identifiers for the active and passive
side of each transaction, precisely signing the trades. Our data is message by message, and
it includes quote changes. This allows us to identify the prevailing quote at the time of each
transaction. A large fraction of dark order transactions in our sample are at the midpoint
and thus incur an effective spread of 0.
To additionally account for the fee that traders of marketable orders pay the exchange,
the “taker” fee, we also compute
taker fee adjusted espreadti = 2qti(pti −mti) + 2× taker feeti. (4)
Colliard and Foucault (2011) refer to this measure as the “cum-fee” spread. In Canada
(as in the U.S.), these fees are sub-penny amounts that accrue per share traded. On the
TSX, these fees can differ by broker, where high-volume brokers pay lower fees. We use
the lowest taker fee, $0.0033 per share, for our computations; the highest is $0.0035. For
orders that execute against a dark order, the taker fee is $0.001.
The change in liquidity provider profits is measured by decomposing the effective spread
into its permanent and transitory components, the price impact and the realized spread,
espreadti = priceimpactti + rspreadti. (5)
The price impact reflects the portion of the transaction costs that is due to the presence
of informed liquidity demanders; a decline in the price impact would indicate a decline in
adverse selection. The realized spread reflects the portion of the transaction costs that is
attributed to liquidity provider revenues. In our analysis we use the five-minute realized
spread, which assumes that liquidity providers are able to close their positions at the
17
midprice five minutes after the trade. The five-minute realized spread is defined as
rspreadti = 2qti(pti −mt+5 min,i), (6)
where pti is the transaction price, mt+5 min,i is the midpoint of the quote 5 minutes after
the t-th trade, and qti is an indicator variable that equals 1 if the trade is buyer-initiated
and −1 if the trade is seller-initiated.
As with effective spreads, we further want to explicitly account for the impact of maker
rebates, and thus compute
maker rebate adjusted rspreadti = 2qti(pti −mt+5 min,i) + 2×maker rebateti, (7)
where maker rebateti is the per share maker fee rebate.15 We use the highest possible rebate,
$0.0031 cents. Dark orders that clear against incoming marketable orders receive no rebate.
V Results on the Introduction of Dark Orders
Stylized Facts of Dark Trading. Table I provides a summary for our sample of firms,
during the period of July 1st to December 31st, 2011, when dark trading was available for
all 229 securities. TSX60 firms are larger and trade more frequently than TSX Completion
firms. As a fraction of orders and trade volume, dark trading in the groups is similar
(3.3% dark trading for TSX60 vs. 2.7%). Dark orders are between 2.4-3 times larger than
standard limit orders. Transaction sizes of dark vs. lit are similar, but this size depends on
the incoming marketable, not-necessarily dark, order. Around 66% of the total Canadian
dollar-volume occurs on the TSX for our sample, consistent with the TSX’s overall market
share of trading as per the regulator’s published statistics.
Quoted depth on the TSX in terms of number of shares is comparable between the
TSX60 and TSX Completion constituents, but since share prices for TSX60 constituents are
15As with taker fees, maker fees depend on the amount of dollar volume that a broker executes on theexchange.
18
larger, the posted dollar-value is larger. Bid-ask spreads for TSX Completion constituents
are larger, both in terms of cents and basis points of the midpoint, and the midpoint is
more volatile compared to TSX60 constituents.
Quoted Liquidity. Figure 2 indicates that the quoted spread increased after the
introduction of dark orders. The average before and after the introduction were 2.18 cents
and 2.44 cents respectively for the TSX60 symbols and 2.82 cents and 2.85 cents for the
TSX Completion symbols. The panel regression results for the change in the quoted spread
are in Panel A of Table III and they confirm that spreads for TSX60 stocks increased after
the introduction of dark orders. The increase is estimated to be between .24 to .32 cents,
depending on the specification and this increase is significant at the 5%-6% level. The
average time-weighted quoted spread before April 01, 2011 for TSX60 constituents was
2.18 cents and thus this marks an increase in spreads of between 11-15%.
Table VII displays the first-stage results from our instrumental variable regression; the
instrument is significant at the 1% level and all tests of poor identification are rejected.
Panel B of Table III displays the results from our second-stage instrumental variable re-
gression. Depending on the specification, they indicate that a 1% increase in the usage
of dark orders would lead to a widening of the spread between .25 and .32 cents. This
observation is consistent with the finding from the first regression as the fraction of dark
trading was about 1% by the end of May.
Finding 1 The introduction of dark orders increased quoted bid-ask-spreads by between
0.24 to 0.32 cents or by 11-15% relative to the average bid-ask spread.
These observations on spreads are consistent with Buti and Rindi (2011) who show theo-
retically that the introduction of fully hidden orders widens quoted bid-ask spreads.
Figure 3 indicates that depth decreased between January and June 2011; however, the
decline was before the introduction of dark orders, and depth was constant afterwards.
Thus even though the panel regression results for the change in the quoted depth from
Table III indicate a decrease in depth, the graph indicates that the drop happened before
the introduction of dark orders. We thus do not attribute this effect to the introduction
19
of dark orders — when we ran the regression for a shorter event window (e.g., starting
mid-February), there was no effect for depth.
Effective Liquidity. In the presence of within-spread dark orders, an increase in
quoted spreads need not be accompanied by an increase in effective spreads. Panel A
Table IV shows, however, that after the introduction of dark orders, effective spreads
increased significantly, by between .17 and .21 cents. In other words, the increase in
quoted spreads dominates the possible reduction in effective spreads that would result
from midpoint executions.
Next, since taker fees for executions against dark orders are lower, an increase in effective
spreads without taker fees need not be accompanied by an increase in effective spreads with
taker fees. Panel A Table IV shows, however, that after the introduction of dark orders
effective spreads with taker fees also increased.
Finding 2 The introduction of dark orders increased effective spreads without and account-
ing for taker fees; the increase is between 0.17 to 0.21 cents.
Other measures. Tables V and VI show that the introduction of dark orders had no
impact on realized spreads plus maker fees, the price impact, nor volume. The fill rate,
i.e. the fraction of order volume that is executed, is also unaffected. We further compute
a measure of mid-price volatility, following Hasbrouck and Saar (2012): we split the day
into 36 10-minute intervals and for each of these we compute the difference between the
highest and the lowest mid-price and scale it by the average mid-price. The average over
these 36 periods is the daily mid-price volatility. For this measure we find no change after
the introduction of dark orders.
Spreads can also be expressed in basis points as a proportion of the prevailing quote
midpoint. We find no significant effect of the introduction on basis point spreads. The
advantage of the proportional spread is that stocks with different price levels may be more
comparable. The caveat is, however, that effects (or non-effects) can be driven entirely by
movements in the underlying stock price (i.e., the denominator in proportional spreads).
For the time from July to December 2011, volume-weighted effective spreads are just under
20
2 cents for TSX60 stocks and 2.6 cents for TSX Completion stocks; so they are different,
but the difference is not major. Scaled by the mid-point, however, TSX60 effective spreads
are 5.4 basis points, and 16.8 for the TSX Completion, due to the, on average, lower prices
that the TSX Completion stocks trade at. As TSX Completion stocks trade at lower dollar
prices, their proportional spreads are systematically more sensitive to price changes.
VI Results for the Intra-Day Usage of Dark Orders
We are presenting the results from a 2-stage instrumental variable regression analysis of
equation (2) where we instrument the percentage of dark executions at time t with the
percentage of dark orders submitted at t− 1. Table VIII indicates that our instrument is
highly significant and that all tests for over-, under- and weak specification are rejected for
all of the estimated specifications.
Regression results for Quoted Liquidity. Table IX presents the results from the
estimations of the intra-day effect of dark trading. Our results indicate that a 1% increase
in dark trading decreases TSX spreads by between .019 to .027 cents, depending on the
model specification. Increases in dark trading also lead to an increase in quoted depth.
Finding 3 An increase in dark trading leads to narrower quoted bid-ask spreads and in-
creased depth.
Taken together, these findings indicate an improvement in quoted liquidity associated with
the contemporaneous existence of dark trading. This finding is notable for two reasons.
First, our results from the introduction of dark trading indicate that quoted spreads in-
creased following the introduction of dark trading. Second, a stylized fact of dark trading
is that the prevailing quoted spreads for trades that involve a dark order exceed those when
the trade involves only lit orders (see Table II). Trades against dark orders can be identified
from the published trades, for instance as within-spread executions.16 Assume that traders
16TMX offers a number of data feeds. The cheapest of these rounds prices and thus midpoint executionsat one cent spreads are not identifiable; data from Thomson Reuters Tick History is based on this feed.Professional traders and major brokerages, however, usually subscribe to the Level 2 feed provides more
21
post wider spreads if they suspect that a dark mid-point order may be present. When an
execution against a dark order is observed, traders learn that (a) there was a dark order
and (b) that dark liquidity has now been reduced. As a result, they may post quotes more
aggressively.
Effective Liquidity. As quoted spreads decline and depth increases, intuition suggests
that effective spreads should decline. Table X confirms this notion. A 1% increase in dark
trading is associated with a 0.45 cents decline in effective spreads. Table XI illustrates
that this effect applies not only when we compute the effective spread based on the local
TSX best bid and offer prices, but also when the effective spread is based on the national
mid-quote. We also compute the effective spreads for orders that involve visible trades on
the passive side. Here, we also find a reduction in the spread that is similar in magnitude
to the reduction in the quoted spread. In other words, the reduction in effective spreads is
not merely caused by the zero-spreads that dark mid-point executions generate.
Finding 4 Dark trading reduces subsequent effective spreads with and without taker fees.
Volume. For TSX data, we generally observe that quoted spreads at transactions are
lower than time-weighted quoted spreads, indicating that executions occur when spreads
are small. As dark trading reduces spreads, one can expect an increase in volume. Table
XIII indeed indicates that this is the case. We further checked whether the ratio of executed
to submitted volume, which we interpret as the fill-rate, is affected by dark trading. Table
XIII shows that fill-rates increase, including for visible-only orders.
Finding 5 Dark trading leads to an increase in volume and fill-rates.
Price Impact and Volatility. Tables XII and XIII outline that there is a reduction
in price impact and volatility following increases in dark trading. One explanation is that
those who have information trade regardless of the spread, but that there are some price-
sensitive uninformed traders that act only once spreads are sufficiently small and who are
thus attracted by the spread reduction following dark executions. This finding is consistent
detailed information.
22
with Comerton-Forde and Putnins (2012), who perform an analysis of the impact of dark
trading on price discovery.
Table XII further outlines a decline in volatility, in line with the reduced average price
impact over the time interval.
Finding 6 Dark trading leads to reduced price impact and lower mid-price volatility.
The effects that we present are statistically significant, but are they economically signif-
icant? We observe a decline in effective spreads, which can be understood as active order
execution costs, by .045 cents for every 1% increase in dark order executions. For perspec-
tive, the exchange fee, i.e. the taker fee minus the maker rebate, that the TSX charges per
transaction is 0.02 cents. Compared to this amount, the observed effect is economically
meaningful.
VII Conclusion
Dark trading is not a new phenomenon. Institutional traders have had access to crossing
networks at least since the 1980s (e.g., ITG’s POSIT has been available since 1987). The
recent development is that all market participants can trade in continuous dark pools and
on public exchanges using fully hidden orders. We study the introduction of fully hidden
orders on the Toronto Stock Exchange. While trading in Canada is fragmented, the TSX
is the main market (with 2/3 of Canadian trading volume) and the introduction of dark
orders thus affects all market participants right away, allowing us to perform an event study.
We study the impact of dark trading from two angles. First, we analyze the impact of
the introduction using a difference-in-differences approach. Consistent with Buti and Rindi
(2011)’s theoretical predictions, we find that the introduction of dark orders increases both
quoted and effective spreads, suggesting that dark trading caused an increase in trading
costs. We don’t find changes in volume or volatility. Second, we study the usage of
dark orders on the intra-day level to understand how dark trading is treated by market
participants. We find that dark trading leads to lower spreads, higher volume, and lower
23
volatility. These facts can be interpreted as two sides of the same coin: upon observing
dark executions, traders may infer that dark liquidity is diminished and thus post quotes
more aggressively.
We caution that our results may understate the potential benefits of having the option
to trade with dark orders, as standard liquidity measures do not capture several advantages.
First, traders may benefit by switching from submitting marketable orders to posting dark
midpoint orders. Dark orders have high fill rates (see Table II), and, compared to market
orders, execute at a price that is superior to the visible quotes. Second, limit orders of
market participants who don’t have the ability to monitor the market at high frequencies
may become stale. A midpoint-pegged order gives these traders a new option to avoid
this pitfall. Third, marketable orders that execute against against dark orders obtain a
better price. While the effective spread accounts for these savings, the uptake of dark
orders just after the introduction may have been too low to reflect this benefit — the
uptake did continue to increase past the time frame that we can use for the difference-in-
difference analysis.
References
Aitken, Michael J, Henk Berkman, and Derek Mak, 2001, The use of undisclosed limit
orders on the australian stock exchange, Journal of Banking and Finance 25, 1589 –
1603.
Anand, Amber, and Daniel G. Weaver, 2004, Can order exposure be mandated?, Journal
of Financial Markets 7, 405 – 426.
Andrews, Donald W. K., and James H. Stock, 2005, Testing for Weak Instruments in
Linear IV Regression (Cambridge University Press).
Bessembinder, Hendrik, Marios Panayides, and Kumar Venkataraman, 2009, Hidden liq-
uidity: An analysis of order exposure strategies in electronic stock markets, Journal of
Financial Economics 94, 361 – 383.
Bessembinder, Hendrik, and Kumar Venkataraman, 2004, Does an electronic stock ex-
change need an upstairs market?, Journal of Financial Economics 73, 3–36.
24
Bloomfield, Robert, Gideon Saar, and Maureen O’Hara, 2012, Hidden liquidity: Some new
light on dark trading, Johnson School Research Paper Series 46-2011 Johnson School,
Cornell University.
Boni, Leslie A, David C. Brown, and J. Chris Leach, 2012, Dark pool
exclusivity matters, working paper University of Colorado Boulder SSRN:
http://ssrn.com/abstract=2055808.
Boulatov, Alex, and Thomas J. George, 2013, Hidden and displayed liquidity in securities
markets with informed liquidity providers, Review of Financial Studies.
Buti, Sabrina, and Barbara Rindi, 2011, Undisclosed orders and optimal submission strate-
gies in a dynamic limit order market, Discussion paper University of Toronto.
, and Ingrid M. Werner, 2011a, Dark pool trading strategies, Working Paper 2010-6
Charles A. Dice Center.
, 2011b, Diving into dark pools, Working Paper No. 2010-03-01 Fisher College of
Business.
Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller, 2011, Robust inference with
multi-way clustering, Journal of Business Economics and Statistics forthcoming.
Colliard, Jean-Edouard, and Thierry Foucault, 2011, Trading fees and efficiency in limit
order markets, working paper HEC Paris.
Comerton-Forde, Carole, and Talis Putnins, 2012, Dark trading and price discovery, work-
ing paper Australian National University.
De Winne, Rudy, and Catherine D’hondt, 2007, Hide-and-seek in the market: Placing and
detecting hidden orders, Review of Finance 11, 663–692.
Degryse, Hans, Frank De Jong, and Vincent Van Kervel, 2011, The impact of dark trading
and visible fragmentation on market quality, Discussion paper No. 2011-026 TILEC.
Frey, Stephan, and Patrik Sandas, 2009, The impact of iceberg orders in limit order books,
working paper University of Virginia.
Hasbrouck, Joel, and Gideon Saar, 2012, Low-latency trading, Research Paper Series No.
35-2010 Johnson School.
Kleibergen, Frank, and Richard Paap, 2006, Generalized reduced rank tests using the
singular value decomposition, Journal of Econometrics 133, 97–126.
25
Moinas, Sophie, 2010, Hidden limit orders and liquidity in order driven markets, Working
paper Toulouse School of Economics.
Nimalendran, Mahendrarajah, and Sugata Ray, 2012, Informational linkages be-
tween dark and lit trading venues, working paper University of Florida
http://ssrn.com/abstract=1880070.
O’Hara, Maureen, and Mao Ye, 2011, Is market fragmentation harming market quality?,
Journal of Financial Economics 100, 459–474.
Pardo, Angel, and Roberto Pascual, 2011, On the hidden side of liquidity, The European
Journal of Finance 1, 1–19.
Petersen, Mitchell A., 2009, Estimating Standard Errors in Finance Panel Data Sets: Com-
paring Approaches, Review of Financial Studies 22, 435–480.
Ready, Mark J., 2009, Determinants of volume in dark pools, working paper University of
Wisconsin http://ssrn.com/abstract=1361234.
Smith, Brian F., D. Alasdair S. Turnbull, and Robert W. White, 2001, Upstairs market
for principal and agency trades: Analysis of adverse information and price effects, The
Journal of Finance 56, 1723–1746.
Thompson, Samuel B., 2011, Simple formulas for standard errors that cluster by both firm
and time, Journal of Financial Economics 99, 1–10.
Weaver, Daniel G., 2011, Internalization and market quality in a fragmented market struc-
ture, Working paper Rutgers University.
Yao, Chen, 2012, Hidden agendas: A study of the impact of concealed orders, Working
paper UICU.
Ye, Mao, 2011, A glimpse into the dark: Price formation, transaction costs, and market
share in the crossing network, working paper University of Illinois at Urbana-Champaign.
Zhu, Haoxiang, 2012, Do dark pools harm price discovery?, Working paper MIT
http://dx.doi.org/10.2139/ssrn.1712173.
26
Table I
Summary Statistics
This table represents summary statistics for the main liquidity and volume variables for our sample ofTSX Composite stocks, split into the groups of TSX60 and TSX Completion constituents. With theexception of market capitalization, which is based on January 2011, all measures are based on daily perstock averages for the second half of 2011. All volume related numbers exclude odd-lot shares; limit orderbook (LOB) figures further exclude dealer crosses, market-on-open, market-on-close, after-hours and specialterms market trades.
TSX60TSXCompletion
TSXComposite
Scaling
Number of firms 53 177 230
av market cap January 2011 21.1 2.6 6.8 (billions)quoted midpoint (in CAD) 39.4 19.4 24.0
Limit order book (LOB) transactions 6.8 1.5 2.7 (thousand)volume 1.4 0.4 0.6 (millions)$-volume 44.9 5.0 14.2 (millions)
Dealer crosses transactions 3.3 1.5 1.9volume 0.3 0.1 0.1 (millions)$-volume 7.4 1.4 2.8 (millions)
Dark Trades (at least one side) transactions 166 27 59volume 1.4 0.1 0.4 (millions)$-volume 0.0 0.0 0.0 (millions)
Orders LOB number 147.7 25.1 53.4 (thousand)volume 56.9 11.6 22.1 (millions)$-volume 1,954 182 590 (millions)
Orders dark number 452 78 164volume 0.5 0.1 0.2 (millions)$-volume 15.5 1.9 5.0 (millions)
Fill-rate LOB orders (volume) 3 4 4 percentFill-rate dark orders (volume) 17 14 15 percent
Trade size in shares LOB 190 239 228dark 214 248 239dealer cross 81,455 64,872 70,094
Order size in shares LOB 380 474 452dark 1,138 1,061 1,080
fragmentation continuous markets 0.53 0.48 0.49% dark of all TSX LOB (volume) 3.3 2.7 2.8 percent% dark of all orders TSX (volume) 1.2 1.2 1.2 percentMarket share TSX $-volume 64.5 66.5 66.0 percent
quoted depth TSX in shares 2.2 2.4 2.4 (thousand)quoted depth TSX in $ 41.9 19.5 24.7 (thousand)quoted spread 2.6 3.4 3.2 centsquoted spread 6.6 19.2 16.3 basis pointseffective spread 2.0 2.6 2.5 centsintra-day volatility 36.0 41.9 40.5 basis points
Table II
Summary Statistics Dark vs. Lit Trading
This table represents summary statistics for several key liquidity and market quality variables split up by dark vs. visible trade for our sample of TSXComposite stocks. All measures are based on daily per stock averages for the second half of 2011. Dark vs. lit is to be understood as follows: we computethe daily average of, for instance, quoted spreads, separately for all executions that involve only lit orders and those that involve dark order on the passiveside of the transaction. We then compute the equal weighted average for all daily, per company, averages. Wilcoxon tests of matched pairs (by firm anddate) for equality of means and medians for the daily lit vs. dark realizations are all rejected at the 1% level except for the equality of means test for tradesize for the TSX Composite, which is rejected at the 5% level of significance.
TSX60 TSX Completion TSX Composite
Visible Dark Diff. Visible Dark Diff. Visible Dark Diff.
Quoted Spread at transaction Average 1.9 2.5 -0.6 2.5 3.4 -0.9 2.4 3.1 -0.7StDev 1.6 2.4 2.7 4.2 2.5 3.8Median 1.2 1.7 1.6 2.0 1.4 2.0
Depth at transaction 42 48 -6 19 22 -3 24 29 -5(in thousands of shares) 33 43 69 107 63 94
31 34 13 12 15 16
Effective Spreads 2.0 0.3 1.7 2.7 0.6 2.1 2.5 0.5 2.01.6 1.5 2.9 3.2 2.7 2.81.3 0.0 1.7 0.0 1.6 0.0
Effective Spreads plus taker fee 2.6 0.5 2.2 3.4 0.8 2.6 3.2 0.7 2.51.6 1.5 2.9 3.2 2.7 2.82.0 0.2 2.4 0.2 2.3 0.2
Realized spread -0.8 0.5 -1.3 -0.7 0.3 -1.0 -0.7 0.3 -1.11.5 9.1 2.4 8.1 2.2 8.4-0.6 0.3 -0.6 0.0 -0.6 0.1
Price Impact 1.4 -0.1 1.5 1.7 0.2 1.6 1.6 0.1 1.61.2 4.7 1.8 4.2 1.7 4.31.0 -0.1 1.2 0.0 1.1 0.0
Fill rate 2.7 17.1 -14.3 4.2 14.3 -10.2 3.8 15.0 -11.22.4 38.1 3.4 53.6 3.3 50.32.1 10.0 3.4 7.3 3.0 7.9
Table III
Panel regressions results for time-weighted quoted liquidityThe table presents the results from a regression where we assess the differential impact of the introduction of dark orders on the treatment group of 53companies from the TSX60 index relative to a control group of 177 companies in the TSX Composite for daily time-weighted quoted spreads (in cents)and the time weighted logarithms of dollar-depth; both are based on TSX based trading. Panel A describes the result from a least square event studyregression and the coefficient of interest is that for the interaction term Introductiont × TSX60i. Panel B describes the results from a 2-stage regressionsto assess the effect of the extent of dark trading on the dependent variable; here, the fraction of dark trading volume of all trading volume on day t instock i is instrumented in the first stage by the interaction variable Introductiont × TSX60i. Control variables are the contemporaneous logarithm of theaverage daily mid-quote, the daily realization of the Canadian volatility index VIXC, and market fragmentation (measure by one minus the HerfindahlIndex of market fragmentation in Canada). We also consider a specification in which we interact the volatility index VIXC with index constituency toallow for differential reactions of the treatment and control group to market-wide volatility. The analysis is is based on daily measures and covers the timespan from January 01, 2011 to May 20, 2011; dark orders were introduced on April 1st for TSX60 symbols and on May 22 for TSX Completion symbols.All regressions include stock fixed effects and fixed effects for the 10-minute interval. Standard errors are double-clustered by firm and time. * indicatessignificance of non-zero correlation at the 10% level, *+ at the 6% level, ** at the 5% level, **+ at the 2% level, and *** at the 1% level.
Panel A: OLS RegressionsDependent variable Time weighted quoted spread Time-weighted Log $-Depth
(1) (2) (3) (4) (1) (2) (3) (4)
Introductiont × TSX60i 0.2475*+ 0.2668** 0.2973** 0.3208** -0.0732**+ -0.0796**+ -0.0882*** -0.0959***(0.1270) (0.1271) (0.1386) (0.1380) (0.0303) (0.0312) (0.0307) (0.0317)
Introductiont 0.0645 -0.0177 0.0531 -0.0304 -0.1590*** -0.1320*** -0.1555*** -0.1281***(0.0556) (0.0580) (0.0550) (0.0576) (0.0172) (0.0169) (0.0171) (0.0168)
Log-Midpointit 1.4836*** 1.5436*** 1.4715*** 1.5307*** 0.2532*** 0.2336*** 0.2569*** 0.2375***(0.2586) (0.2628) (0.2589) (0.2629) (0.0754) (0.0766) (0.0755) (0.0767)
VIXCt 0.0631*** 0.0639*** -0.0326*** -0.0329***(0.0163) (0.0157) (0.0050) (0.0046)
Fragmentationit 1.9512*** 1.9588*** -0.6400*** -0.6424***(0.3285) (0.3281) (0.0550) (0.0549)
VIXCt × TSX60i 0.1119*** 0.1168*** -0.0473*** -0.0489***(0.0248) (0.0244) (0.0081) (0.0080)
VIXCt × TSX Completioni 0.0484*** 0.0480*** -0.0282*** -0.0281***(0.0171) (0.0165) (0.0045) (0.0042)
Observations 21,607 21,607 21,607 21,607 21,607 21,607 21,607 21,607adjusted R-squared 0.862 0.865 0.862 0.865 0.799 0.806 0.8 0.806
Panel B: 2-stage IV EstimationDependent variable Time weighted quoted spread Time-weighted Log $-Depth
(1) (2) (3) (4) (1) (2) (3) (4)
% dark TSXit 0.2511** 0.2704** 0.2949** 0.3179**+ -0.0743** -0.0806**+ -0.0875*** -0.0951***(0.1257) (0.1265) (0.1359) (0.1363) (0.0332) (0.0346) (0.0337) (0.0353)
Introduction 0.0633 -0.0179 0.0532 -0.029 -0.1586*** -0.1319*** -0.1555*** -0.1286***(0.0558) (0.0580) (0.0550) (0.0574) (0.0173) (0.0169) (0.0171) (0.0167)
Log-Midpointit 1.4723*** 1.5306*** 1.4599*** 1.5172*** 0.2566*** 0.2374*** 0.2603*** 0.2415***(0.2585) (0.2620) (0.2596) (0.2629) (0.0765) (0.0776) (0.0768) (0.0779)
VIXCt 0.0614*** 0.0621*** -0.0321*** -0.0323***(0.0163) (0.0157) (0.0050) (0.0047)
Fragmentationit 1.9259*** 1.9281*** -0.6325*** -0.6332***(0.3271) (0.3269) (0.0557) (0.0557)
VIXCt × TSX60i 0.1034*** 0.1076*** -0.0448*** -0.0462***(0.0249) (0.0251) (0.0084) (0.0084)
VIXCt × TSX Completioni 0.0483*** 0.0479*** -0.0282*** -0.0280***(0.0171) (0.0165) (0.0045) (0.0042)
Table IV
Panel regressions results for volume-weighted effective spreadsThe table presents the results from a regression where we assess the differential impact of the introduction of dark orders on the treatment group of 53companies from the TSX60 index relative to a control group of 177 companies in the TSX Composite for daily volume-weighted effective spreads withoutand with taker fees; both are measured in cents and are based on TSX based trading. Panel A describes the result from a least square event studyregression and the coefficient of interest is that for the interaction term Introductiont × TSX60i. In contrast to Table III, this table does not contain a2-stage regressions to assess the effect of the extent of dark trading on the dependent variable, because such a regression would be misleading: Table IIIshows that quoted spreads increased after the introduction of dark orders. However, since dark trades execute at the midpoint, and increase in the extentof dark trading, ceteris paribus, would mechanically lower effective spreads. Thus, there are two opposing forces, the second of which, however, cannot berepresented adequately in the IV regression. Control variables are the contemporaneous logarithm of the average daily mid-quote, the daily realizationof the Canadian volatility index VIXC, and market fragmentation (measure by one minus the Herfindahl Index of market fragmentation in Canada). Wealso consider a specification in which we interact the volatility index VIXC with index constituency to allow for differential reactions of the treatment andcontrol group to market-wide volatility. The analysis is is based on daily measures and covers the time span from January 01, 2011 to May 20, 2011; darkorders were introduced on April 1st for TSX60 symbols and on May 22 for TSX Completion symbols. All regressions include stock fixed effects and fixedeffects for the 10-minute interval. Standard errors are double-clustered by firm and time. * indicates significance of non-zero correlation at the 10% level,*+ at the 6% level, ** at the 5% level, **+ at the 2% level, and *** at the 1% level.
Panel A: OLS RegressionsDependent variable Effective spread Effective spread plus taker fee
(1) (2) (3) (4) (1) (2) (3) (4)
Introductiont × TSX60i 0.1594* 0.1711** 0.1904** 0.2046** 0.1547* 0.1664*+ 0.1856** 0.1998**(0.0859) (0.0868) (0.0919) (0.0922) (0.0859) (0.0867) (0.0918) (0.0921)
Introductiont -0.0714 -0.1210**+ -0.0785 -0.1289**+ -0.0714 -0.1210**+ -0.0785 -0.1289**+(0.0498) (0.0507) (0.0495) (0.0503) (0.0497) (0.0507) (0.0495) (0.0503)
Log-Midpointit 1.1067*** 1.1429*** 1.0992*** 1.1349*** 1.1065*** 1.1427*** 1.0990*** 1.1347***(0.1989) (0.2038) (0.1994) (0.2042) (0.1989) (0.2038) (0.1994) (0.2042)
Canadian VIXCt 0.0411*** 0.0416*** 0.0411*** 0.0416***(0.0122) (0.0120) (0.0122) (0.0120)
Fragmentationit 1.1770*** 1.1818*** 1.1766*** 1.1813***(0.2569) (0.2564) (0.2569) (0.2564)
VIXCt × TSX60i 0.0715*** 0.0745*** 0.0714*** 0.0744***(0.0165) (0.0163) (0.0165) (0.0164)
VIXCt × TSX Completioni 0.0320**+ 0.0317**+ 0.0320**+ 0.0317**+(0.0136) (0.0134) (0.0136) (0.0134)
Observations 21,607 21,607 21,607 21,607 21,607 21,607 21,607 21,607adjusted R-squared 0.789 0.791 0.789 0.791 0.788 0.79 0.789 0.791
Table V
Panel regressions results for volume-weighted realized spreads and price impactsThe table presents the results from a regression where we assess the differential impact of the introduction of dark orders on the treatment group of 53companies from the TSX60 index relative to a control group of 177 companies in the TSX Composite for daily volume-weighted 5-minute realized spreadswith maker rebate and volume-weighted 5-minute price impacts; both are measured in cents and are based on TSX based trading. Panel A describes theresult from a least square event study regression and the coefficient of interest is that for the interaction term Introductiont × TSX60i. Panel B describesthe results from a 2-stage regressions to assess the effect of the extent of dark trading on the dependent variable; here, the fraction of dark trading volumeof all trading volume on day t in stock i is instrumented in the first stage by the interaction variable Introductiont × TSX60i. Control variables arethe contemporaneous logarithm of the average daily mid-quote, the daily realization of the Canadian volatility index VIXC, and market fragmentation(measure by one minus the Herfindahl Index of market fragmentation in Canada). We also consider a specification in which we interact the volatilityindex VIXC with index constituency to allow for differential reactions of the treatment and control group to market-wide volatility. The analysis is isbased on daily measures and covers the time span from January 01, 2011 to May 20, 2011; dark orders were introduced on April 1st for TSX60 symbolsand on May 22 for TSX Completion symbols. All regressions include stock fixed effects and fixed effects for the 10-minute interval. Standard errors aredouble-clustered by firm and time. * indicates significance of non-zero correlation at the 10% level, *+ at the 6% level, ** at the 5% level, **+ at the 2%level, and *** at the 1% level.
Panel A: OLS RegressionsDependent variable realized spread plus maker rebate 5-minute price impact
(1) (2) (3) (4) (1) (2) (3) (4)
Introductiont × TSX60i 0.0150 0.0146 0.0197 0.0193 0.0691 0.0751 0.0821 0.0894(0.1001) (0.1000) (0.1051) (0.1050) (0.0598) (0.0592) (0.0650) (0.0643)
Introductiont -0.0675 -0.0659 -0.0686 -0.067 -0.002 -0.0276 -0.005 -0.0309(0.0557) (0.0558) (0.0559) (0.0560) (0.0303) (0.0301) (0.0300) (0.0300)
Log-Midpointit -0.7744*** -0.7755*** -0.7755*** -0.7766*** 0.9404*** 0.9591*** 0.9372*** 0.9556***(0.2344) (0.2335) (0.2346) (0.2337) (0.1527) (0.1520) (0.1526) (0.1519)
VIXCt -0.0250* -0.0250* 0.0330*** 0.0333***(0.0144) (0.0144) (0.0100) (0.0099)
Fragmentationit -0.0376 -0.0369 0.6070*** 0.6091***(0.1605) (0.1607) (0.1269) (0.1270)
VIXCt × TSX60i -0.0203 -0.0204 0.0458*** 0.0474***(0.0166) (0.0167) (0.0128) (0.0127)
VIXCt × TSX Completioni -0.0264 -0.0264 0.0292*** 0.0291***(0.0172) (0.0172) (0.0112) (0.0110)
Observations 21,607 21,607 21,607 21,607 21,607 21,607 21,607 21,607adjusted R-squared 0.0831 0.0831 0.0831 0.083 0.507 0.508 0.507 0.508
Panel B: 2-stage IV EstimationDependent variable realized spread plus maker rebate 5-minute price impact
(1) (2) (3) (4) (1) (2) (3) (4)
0.0152 0.0148 0.0196 0.0191 0.07 0.0761 0.0814 0.0886% dark TSXit (0.1023) (0.1021) (0.1051) (0.1049) (0.0579) (0.0571) (0.0620) (0.0612)
-0.0675 -0.0659 -0.0686 -0.0669 -0.0023 -0.0276 -0.005 -0.0305Introductiont (0.0560) (0.0559) (0.0559) (0.0558) (0.0304) (0.0301) (0.0300) (0.0298)
-0.7751*** -0.7762*** -0.7763*** -0.7775*** 0.9373*** 0.9554*** 0.9340*** 0.9519***Log-Midpointit (0.2343) (0.2334) (0.2343) (0.2334) (0.1540) (0.1534) (0.1543) (0.1537)
-0.0251* -0.0251* 0.0325*** 0.0328***VIXCt (0.0143) (0.0143) (0.0101) (0.0099)
-0.039 -0.0388 0.5999*** 0.6005***Fragmentationit (0.1611) (0.1612) (0.1270) (0.1270)
-0.0209 -0.021 0.0435*** 0.0448***VIXCt × TSX60i (0.0154) (0.0155) (0.0128) (0.0129)
-0.0264 -0.0264 0.0291*** 0.0290***VIXCt × TSX Completioni (0.0172) (0.0172) (0.0112) (0.0110)
Table VI
Panel regressions results for volatility, volume and fill-ratesThe table presents the results from a regression where we assess the differential impact of the introduction of dark orders on the treatment group of 53companies from the TSX60 index relative to a control group of 177 companies in the TSX Composite for the daily 10-minute midpoint volatility (highestquoted midpoint minus lowest quoted midpoint, scaled by the average quoted midpoint, measured in basis points), the logarithms of dollar volume, andthe fill-rate (the fraction of traded volume of all submitted order volume). Panel A describes the result from a least square event study regression and thecoefficient of interest is that for the interaction term Introductiont × TSX60i. Panel B describes the results from a 2-stage regressions to assess the effect ofthe extent of dark trading on the dependent variable; here, the fraction of dark trading volume of all trading volume on day t in stock i is instrumented inthe first stage by the interaction variable Introductiont × TSX60i. Control variables are the contemporaneous logarithm of the average daily mid-quote, thedaily realization of the Canadian volatility index VIXC, and market fragmentation (measure by one minus the Herfindahl Index of market fragmentation inCanada). We also consider a specification in which we interact the volatility index VIXC with index constituency to allow for differential reactions of thetreatment and control group to market-wide volatility. The analysis is is based on daily measures and covers the time span from January 01, 2011 to May20, 2011; dark orders were introduced on April 1st for TSX60 symbols and on May 22 for TSX Completion symbols. All regressions include stock fixedeffects and fixed effects for the 10-minute interval. Standard errors are double-clustered by firm and time. * indicates significance of non-zero correlationat the 10% level, *+ at the 6% level, ** at the 5% level, **+ at the 2% level, and *** at the 1% level.
Panel A: OLS RegressionsDependent variable midpoint volatility Log $-Volume Fill rate
(1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4)
Introductiont × TSX60i 0.4901 0.4927 0.5721 0.5754 0.0438 0.0278 0.0549 0.0355 -0.0053 -0.0734 -0.0899 -0.1728(0.6719) (0.6732) (0.7157) (0.7175) (0.0393) (0.0376) (0.0390) (0.0372) (0.2204) (0.2124) (0.2245) (0.2128)
Introductiont 1.6050** 1.5932** 1.5859** 1.5735*+ -0.1812*** -0.1131*** -0.1837*** -0.1149*** -0.0697 0.2205 -0.0503 0.2438(0.7989) (0.8096) (0.7996) (0.8107) (0.0316) (0.0298) (0.0314) (0.0297) (0.2087) (0.1927) (0.2096) (0.1931)
Log-Midpointit 2.4092 2.4217 2.3889 2.4017 1.3515*** 1.3018*** 1.3488*** 1.3000*** -0.4743 -0.6859 -0.4537 -0.6623(4.0781) (4.0854) (4.0764) (4.0836) (0.1559) (0.1636) (0.1559) (0.1637) (0.6924) (0.7648) (0.6934) (0.7653)
VIXCt 1.8565*** 1.8567*** 0.0210** 0.0203**+ 0.0879**+ 0.0849**+(0.3072) (0.3071) (0.0091) (0.0086) (0.0364) (0.0364)
Fragmentationit 0.2874 0.2985 -1.6161*** -1.6151*** -6.8906*** -6.9046***(1.4807) (1.4812) (0.0854) (0.0854) (1.0248) (1.0238)
VIXCt × TSX60i 1.9366*** 1.9373*** 0.0319*** 0.0278**+ 0.005 -0.0124(0.3084) (0.3086) (0.0119) (0.0114) (0.0307) (0.0299)
VIXCt × TSX Completioni 1.8320*** 1.8320*** 0.0178*+ 0.0180** 0.1129*** 0.1142***(0.3265) (0.3264) (0.0093) (0.0088) (0.0428) (0.0427)
Observations 21,306 21,306 21,306 21,306 21,607 21,607 21,607 21,607 21,607 21,607 21,607 21,607adjusted R-squared 0.544 0.544 0.544 0.544 0.867 0.879 0.867 0.879 0.509 0.537 0.509 0.537
Panel B: 2-stage IV EstimationDependent variable midpoint volatility Log $-Volume Fill rate
(1) (2) (3) (4) (1) (2) (3) (4) (1) (2) (3) (4)
0.4972 0.4994 0.5675 0.5701 0.0444 0.0282 0.0544 0.0351 -0.0054 -0.0744 -0.0891 -0.1712% dark TSXit (0.6845) (0.6852) (0.7144) (0.7154) (0.0412) (0.0387) (0.0407) (0.0378) (0.2235) (0.2148) (0.2218) (0.2107)
1.6025** 1.5926** 1.5860** 1.5759*+ -0.1814*** -0.1132*** -0.1837*** -0.1148*** -0.0697 0.2205 -0.0503 0.2431Introductiont (0.8004) (0.8100) (0.7996) (0.8094) (0.0316) (0.0298) (0.0314) (0.0297) (0.2096) (0.1928) (0.2095) (0.1923)
2.3865 2.3968 2.3661 2.3763 1.3495*** 1.3005*** 1.3466*** 1.2985*** -0.4741 -0.6824 -0.4502 -0.655Log-Midpointit (4.0871) (4.0948) (4.0850) (4.0926) (0.1564) (0.1642) (0.1565) (0.1643) (0.6942) (0.7662) (0.6954) (0.7666)
1.8531*** 1.8532*** 0.0207** 0.0201**+ 0.0880**+ 0.0854**VIXCt (0.3072) (0.3072) (0.0092) (0.0086) (0.0367) (0.0367)
0.2394 0.242 -1.6188*** -1.6185*** -6.8836*** -6.8881***Fragmentationit (1.4814) (1.4815) (0.0859) (0.0860) (1.0314) (1.0306)
1.9202*** 1.9208*** 0.0303**+ 0.0268**+ 0.0075 -0.0075VIXCt × TSX60i (0.3063) (0.3064) (0.0120) (0.0115) (0.0309) (0.0312)
1.8318*** 1.8318*** 0.0177*+ 0.0180** 0.1129*** 0.1142***VIXCt × TSX Completioni (0.3265) (0.3264) (0.0093) (0.0087) (0.0428) (0.0428)
Table VII
First stage of the IV regression on the introduction of dark orders
The table presents the results from the first-stage regressions that underlie Panels B in Tables III, V, andVI where we instrument the fraction of dark trading volume of all trading volume on day t in stock i bythe interaction variable Introductiont × TSX60i. Control variables are the contemporaneous logarithmof the average daily mid-quote, the daily realization of the Canadian volatility index VIXC, and marketfragmentation (measure by one minus the Herfindahl Index of market fragmentation in Canada). We alsoconsider a specification in which we interact the volatility index VIXC with index constituency to allow fordifferential reactions of the treatment and control group to market-wide volatility. The analysis is is basedon daily measures and covers the time span from January 01, 2011 to May 20, 2011; dark orders wereintroduced on April 1st for TSX60 symbols and on May 22 for TSX Completion symbols. All regressionsinclude stock fixed effects and fixed effects for the 10-minute interval. Standard errors are double-clusteredby firm and time. * indicates significance of non-zero correlation at the 10% level, *+ at the 6% level,** at the 5% level, **+ at the 2% level, and *** at the 1% level. This table represents the results fromthe first stage of the IV regressions where we instrumented the fraction of dark trading on the TSX withthe introduction-TSX60 dummy. We include the Kleibergen and Paap (2006) Wald statistic of under-identification, which, in our specification is χ2(1) distributed, and the Kleibergen and Paap (2006) statisticfor weak identification (following the Andrews and Stock (2005) critical values; for our specification, the10% maximal IV size critical value is 16.38). All regressions include firm fixed effects, and standard errorsare double-clustered by firm and date.
instrumented variable: % dark TSX trading (1) (2) (3) (4)
Introductiont × TSX60i 0.9858*** 0.9868*** 1.0082*** 1.0094***(0.1473) (0.1472) (0.1403) (0.1403)
Introductiont 0.0049 0.0009 -0.0003 -0.0044(0.0060) (0.0063) (0.0008) (0.0032)
Log-Midpointit 0.045 0.0479 0.0395 0.0425(0.1056) (0.1059) (0.1051) (0.1053)
VIXCt 0.0069 0.0069(0.0069) (0.0069)
Fragmentationit 0.0936 0.0968(0.0706) (0.0713)
VIXCt × TSX60i 0.0288 0.029(0.0292) (0.0292)
VIXCt × TSX Completioni 0.0002 0.0002(0.0007) (0.0007)
Observations 21,607 21,607 21,607 21,607F(4,93) 15.02 12.06 12.06 10.10Kleibergen and Paap (2006) under-identification 21.00 21.01 21.08 21.09Kleibergen and Paap (2006) weak identification 44.80 44.93 51.62 51.79
Table VIII
First stage of the IV regression on the intra-day impact of trading in dark orders
This table represents the results from the first stage of the IV regressions where we instrumented thefraction of dark trading volume on the TSX, % dark tradingit, by the fraction of dark order volume inthe preceding time period, % dark ordersit−1. We include the Kleibergen and Paap (2006) Wald statisticof under-identification, which, in our specification is χ2(1) distributed, and the Kleibergen and Paap(2006) statistic for weak identification (following the Andrews and Stock (2005) critical values; for ourspecification, the 10% maximal IV size critical value is 16.38). Control variables are the daily realizationof the Canadian volatility index VIXC, the average order size in the stock in the preceding week, thecontemporaneous realizations of the logarithm of the time-weighted average mid-quote (over the 10-minuteinterval), market fragmentation (measure by one minus the Herfindahl Index of market fragmentationin Canada), the logarithm of dollar-volume, the fill-rate (the fraction of executed volume of all ordervolume), the 10-minute price volatility (highest quoted midpoint minus lowest quoted midpoint, scaled bythe average quoted midpoint, measured in basis points), and the contemporaneous average over all othersymbols for log-$-volume, trade size and volatility. The sample consists of 229 companies that are in theTSX Composite Index and it covers the time span from July 01, 2011 to December 30, 2011. All regressionsinclude stock fixed effects and fixed effects for the 10-minute interval. Standard errors are double-clusteredby firm and time. * indicates significance of non-zero correlation at the 10% level, *+ at the 6% level, **at the 5% level, **+ at the 2% level, and *** at the 1% level.
instrumented variable: % dark TSX tradingMain instrument: % dark dark ordersit−1
(1) (2) (3) (4) (5)
% dark ordersit−1 0.5635*** 0.5642*** 0.5633*** 0.5606*** 0.5633***(0.021) (0.021) (0.021) (0.021) (0.021)
VIXCd 0.0573*** 0.0643*** 0.0646*** 0.0626*** 0.0638***(0.006) (0.007) (0.007) (0.007) (0.007)
log mid-quoteit 1.0042*** 0.9954*** 0.8591** 0.9798***(0.350) (0.349) (0.377) (0.349)
Fragmentationit 1.9284*** 1.9269*** 2.1267*** 1.9271***(0.145) (0.146) (0.160) (0.146)
log $-volumeit 0.0113(0.038)
Average order size one week agoi 0.0007*+(0.000)
Fillrateit 0.0011(0.003)
Volatilityit -0.0113***(0.001)
Average log-$-volume in all other symbolsit 0.0000(0.000)
Average trade size in all other symbolsit -17.0212(12.410)
Average volatility in all other symbolsit -0.0000***(0.000)
Observations 715,212 715,212 715,212 715,159 715,212adjusted R-squared 0.1 0.099 0.1 0.102 0.1F 32.18 25.42 32.04 30.56 30.21Kleibergen and Paap (2006) under-identification 106.4 106.6 106.5 106.8 106.5Kleibergen and Paap (2006) weak identification 707.2 704.5 704.5 703.4 704.5
Table IX
IV regression on the intra-day impact of trading in dark orders on quoted liquidityThe table presents the regression results from the second stage of an instrumental variable regressions of time-weighted quoted spreads (in cents) andlog-$-depth (both on the TSX) on the fraction of volume that is traded with dark orders and a number of control variables. In the first stage, the variableof interest, % dark tradingit, is instrumented by the fraction of dark orders in the preceding time period, % dark ordersit−1. Control variables are thedaily realization of the Canadian volatility index VIXC, the average order size in the stock in the preceding week, the contemporaneous realizations ofthe logarithm of the time-weighted average mid-quote (over the 10-minute interval), market fragmentation (measure by one minus the Herfindahl Indexof market fragmentation in Canada), the logarithm of dollar-volume, the fill-rate (the fraction of executed volume of all order volume), the 10-minuteprice volatility (highest quoted midpoint minus lowest quoted midpoint, scaled by the average quoted midpoint, measured in basis points), and thecontemporaneous average over all other symbols for log-$-volume, trade size and volatility. All regressions include stock fixed effects and fixed effects forthe 10-minute interval. The sample consists of 229 companies that are in the TSX Composite Index and it covers the time span from July 01, 2011 toDecember 30, 2011. Standard errors are double-clustered by firm and time. * indicates significance of non-zero correlation at the 10% level, *+ at the 6%level, ** at the 5% level, **+ at the 2% level, and *** at the 1% level.
Second stage of the intra-day IV estimationMain instrument: % dark ordersit−1
Dependent variable time-weighted quoted spread time-weighted log-$-depth
(1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
% dark tradingit -0.0267*** -0.0255*** -0.0270*** -0.0198*** -0.0269*** 0.0072*** 0.0066*** 0.0070*** 0.0048**+ 0.0070***(0.005) (0.005) (0.005) (0.005) (0.005) (0.002) (0.002) (0.002) (0.002) (0.002)
VIXCd 0.0350*** 0.0410*** 0.0413*** 0.0463*** 0.0430*** -0.0102*** -0.0065*** -0.0066*** -0.0078*** -0.0065***(0.004) (0.004) (0.004) (0.005) (0.004) (0.001) (0.002) (0.002) (0.002) (0.002)
log mid-quoteit 0.8793*** 0.8724*** 1.2101*** 0.9070*** 0.4887*** 0.4904*** 0.4623*** 0.4922***(0.177) (0.187) (0.186) (0.187) (0.159) (0.158) (0.166) (0.159)
Fragmentationit 1.8278*** 1.8270*** 1.1567*** 1.8266*** -0.4662*** -0.4666*** -0.2421*** -0.4666***(0.191) (0.191) (0.139) (0.191) (0.018) (0.018) (0.013) (0.018)
log $-volumeit -0.2412*** 0.1036***(0.027) (0.005)
Average order size one week agoi 0 0.0003***(0.000) (0.000)
Fillrateit -0.0186*** 0.0063***(0.004) (0.001)
Volatilityit 0.0129*** -0.0027***(0.002) (0.000)
Average log-$-volume in all other symbolsit 0.0000 0.0000(0.000) (0.000)
Average trade size in all other symbolsit 37.7414*** 1.8790*(6.141) (1.132)
Average volatility in all other symbolsit 0.0000*** -0.0000***(0.000) (0.000)
Observations 715,212 715,212 715,212 715,159 715,212 715,212 715,212 715,212 715,159 715,212adjusted R-squared 0.024 0.01 0.025 0.082 0.026 0.03 0.019 0.038 0.126 0.038
Table X
IV regression on the intra-day impact of trading in dark orders on effective trading costsThe table presents the regression results from the second stage of an instrumental variable regressions of volume weighted effective spreads (in cents) andvolume weighted effective spreads plus taker fees (in cents) (both on the TSX) on the fraction of volume that is traded with dark orders and a number ofcontrol variables. In the first stage, the variable of interest, % dark tradingit, is instrumented by the fraction of dark orders in the preceding time period, %dark ordersit−1. Control variables are the daily realization of the Canadian volatility index VIXC, the average order size in the stock in the preceding week,the contemporaneous realizations of the logarithm of the time-weighted average mid-quote (over the 10-minute interval), market fragmentation (measureby one minus the Herfindahl Index of market fragmentation in Canada), the logarithm of dollar-volume, the fill-rate (the fraction of executed volume of allorder volume), the 10-minute price volatility (highest quoted midpoint minus lowest quoted midpoint, scaled by the average quoted midpoint, measured inbasis points), and the contemporaneous average over all other symbols for log-$-volume, trade size and volatility. All regressions include stock fixed effectsand fixed effects for the 10-minute interval. The sample consists of 229 companies that are in the TSX Composite Index and it covers the time span fromJuly 01, 2011 to December 30, 2011. Standard errors are double-clustered by firm and time. * indicates significance of non-zero correlation at the 10%level, *+ at the 6% level, ** at the 5% level, **+ at the 2% level, and *** at the 1% level.
Second stage of the intra-day IV estimationMain instrument: % dark ordersit−1
Dependent variable Effective spread Effective spread plus taker fee
(1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
% dark tradingit -0.0455*** -0.0446*** -0.0456*** -0.0417*** -0.0456*** -0.0460*** -0.0450*** -0.0461*** -0.0422*** -0.0461***(0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004)
VIXCd 0.0178*** 0.0211*** 0.0214*** 0.0233*** 0.0230*** 0.0178*** 0.0211*** 0.0213*** 0.0233*** 0.0229***(0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003)
log mid-quoteit 0.4943*** 0.4891*** 0.6810*** 0.5212*** 0.4930*** 0.4879*** 0.6800*** 0.5200***(0.123) (0.130) (0.131) (0.130) (0.123) (0.130) (0.131) (0.130)
Fragmentationit 1.3689*** 1.3685*** 1.0412*** 1.3681*** 1.3684*** 1.3680*** 1.0404*** 1.3676***(0.135) (0.135) (0.109) (0.135) (0.135) (0.135) (0.109) (0.135)
log $-volumeit -0.0683*** -0.0685***(0.014) (0.014)
Average order size one week agoi 0 0(0.000) (0.000)
Fillrateit -0.0110*** -0.0110***(0.002) (0.002)
Volatilityit 0.0083*** 0.0083***(0.001) (0.001)
Average log-$-volume in all other symbolsit 0.0000 0.0000(0.000) (0.000)
Average trade size in all other symbolsit 35.0160*** 35.0663***(4.945) (4.945)
Average volatility in all other symbolsit 0.0000*** 0.0000***(0.000) (0.000)
Observations 715,212 715,212 715,212 715,159 715,212 715,212 715,212 715,212 715,159 715,212adjusted R-squared 0.039 0.029 0.04 0.064 0.041 0.04 0.03 0.04 0.065 0.041
Table XI
IV regression on the intra-day impact of dark order trading on NBBO and lit-only effective spreadsThe table presents the regression results from the second stage of an instrumental variable regressions of volume weighted effective spreads (in cents, ontheTSX) and the volume weighted effective spreads (in cents) based on the national mid-quote on the fraction of volume that is traded with dark ordersand a number of control variables. In the first stage, the variable of interest, % dark tradingit, is instrumented by the fraction of dark orders in thepreceding time period, % dark ordersit−1. Control variables are the daily realization of the Canadian volatility index VIXC, the average order size inthe stock in the preceding week, the contemporaneous realizations of the logarithm of the time-weighted average mid-quote (over the 10-minute interval),market fragmentation (measure by one minus the Herfindahl Index of market fragmentation in Canada), the logarithm of dollar-volume, the fill-rate (thefraction of executed volume of all order volume), the 10-minute price volatility (highest quoted midpoint minus lowest quoted midpoint, scaled by theaverage quoted midpoint, measured in basis points), and the contemporaneous average over all other symbols for log-$-volume, trade size and volatility.The sample consists of 229 companies that are in the TSX Composite Index and it covers the time span from July 01, 2011 to December 30, 2011. Allregressions include stock fixed effects and fixed effects for the 10-minute interval. Standard errors are double-clustered by firm and time. * indicatessignificance of non-zero correlation at the 10% level, *+ at the 6% level, ** at the 5% level, **+ at the 2% level, and *** at the 1% level.
Second stage of the intra-day IV estimationMain instrument: % dark ordersit−1
Dependent variable volume-weighted effective spread – only lit trading volume-weighted effective spread based on the NBBO
(1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
% dark tradingit -0.0215*** -0.0204*** -0.0217*** -0.0173*** -0.0217*** -0.0434*** -0.0427*** -0.0435*** -0.0402*** -0.0435***(0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) (0.004)
VIXCd 0.0185*** 0.0217*** 0.0220*** 0.0237*** 0.0236*** 0.0194*** 0.0225*** 0.0227*** 0.0241*** 0.0241***(0.003) (0.003) (0.003) (0.003) (0.003) (0.002) (0.003) (0.003) (0.003) (0.003)
log mid-quoteit 0.4824*** 0.4774*** 0.6625*** 0.5096*** 0.4583*** 0.4543*** 0.6050*** 0.4819***(0.123) (0.131) (0.131) (0.131) (0.112) (0.118) (0.119) (0.118)
Fragmentationit 1.4052*** 1.4048*** 1.0758*** 1.4044*** 1.0645*** 1.0641*** 0.7894*** 1.0638***(0.139) (0.139) (0.112) (0.138) (0.109) (0.109) (0.090) (0.109)
log $-volumeit -0.0625*** -0.0492***(0.014) (0.011)
Average order size one week agoi 0 0(0.000) (0.000)
Fillrateit -0.0113*** -0.0099***(0.002) (0.002)
Volatilityit 0.0084*** 0.0070***(0.001) (0.001)
Average log-$-volume in all other symbolsit 0.0000 0.0000(0.000) (0.000)
Average trade size in all other symbolsit 35.1427*** 30.1122***(4.940) (4.157)
Average volatility in all other symbolsit 0.0000*** 0.0000***(0.000) (0.000)
Observations 713,413 713,413 713,413 713,363 713,413 715,212 715,212 715,212 715,159 715,212adjusted R-squared 0.016 0.005 0.016 0.042 0.017 0.041 0.034 0.042 0.065 0.043
Table XII
IV regression on the intra-day impact of trading in dark orders on price impact and realized spreads + maker feesThe table presents the regression results from the second stage of an instrumental variable regressions of volume weighted 5-minute realized spreads plusmaker rebate (in cents) and the volume weighted 5-minute price impact (in cents) (both on the TSX) on the fraction of volume that is traded with darkorders and a number of control variables. In the first stage, the variable of interest, % dark tradingit, is instrumented by the fraction of dark orders inthe preceding time period, % dark ordersit−1. Control variables are the daily realization of the Canadian volatility index VIXC, the average order size inthe stock in the preceding week, the contemporaneous realizations of the logarithm of the time-weighted average mid-quote (over the 10-minute interval),market fragmentation (measure by one minus the Herfindahl Index of market fragmentation in Canada), the logarithm of dollar-volume, the fill-rate (thefraction of executed volume of all order volume), the 10-minute price volatility (highest quoted midpoint minus lowest quoted midpoint, scaled by theaverage quoted midpoint, measured in basis points), and the contemporaneous average over all other symbols for log-$-volume, trade size and volatility.The sample consists of 229 companies that are in the TSX Composite Index and it covers the time span from July 01, 2011 to December 30, 2011. Allregressions include stock fixed effects and fixed effects for the 10-minute interval. Standard errors are double-clustered by firm and time. * indicatessignificance of non-zero correlation at the 10% level, *+ at the 6% level, ** at the 5% level, **+ at the 2% level, and *** at the 1% level.
Second stage of the intra-day IV estimationMain instrument: % dark ordersit−1
Dependent variable 5-minute price impact 5-minute realized plus maker rebate
(1) (2) (3) (4) (5) (1) (2) (3) (4) (5)
% dark tradingit -0.0246*** -0.0242*** -0.0247*** -0.0175*** -0.0246*** 0.0030 0.0031 0.0030 -0.0074 0.0030(0.002) (0.002) (0.002) (0.002) (0.002) (0.005) (0.005) (0.005) (0.005) (0.005)
VIXCd 0.0125*** 0.0142*** 0.0143*** 0.0184*** 0.0158*** -0.0073*** -0.0074*** -0.0074*** -0.0136*** -0.0086***(0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.003) (0.003) (0.002)
log mid-quoteit 0.2532*** 0.2508*** 0.6685*** 0.2795*** -0.0138 -0.0142 -0.6574*** -0.0394(0.082) (0.084) (0.087) (0.084) (0.110) (0.110) (0.126) (0.109)
Fragmentationit 0.6336*** 0.6334*** 0.1847*** 0.6331*** 0.101 0.101 0.6708*** 0.1012(0.077) (0.077) (0.055) (0.077) (0.077) (0.077) (0.083) (0.077)
log $-volumeit -0.0505*** 0.0323(0.015) (0.022)
Average order size one week agoi 0 0(0.000) (0.000)
Fillrateit -0.0096*** 0.0083***(0.002) (0.002)
Volatilityit 0.0181*** -0.0279***(0.002) (0.002)
Average log-$-volume in all other symbolsit 0.0000 0.0000(0.000) (0.000)
Average trade size in all other symbolsit 31.3064*** -27.5291***(4.767) (7.196)
Average volatility in all other symbolsit -0.0000*** 0.0000***(0.000) (0.000)
Observations 715,212 715,212 715,212 715,159 715,212 715,212 715,212 715,212 715,159 715,212adjusted R-squared 0.012 0.01 0.012 0.077 0.012 0.001 0.001 0.001 0.04 0.001
Table XIII
IV regression on the intra-day impact of trading in dark orders on volatility, volume and fill ratesThe table presents the regression results from the second stage of an instrumental variable regressions of 10-minute mid-quote volatility, log-dollar-volume,the fill-rate for all orders (the fraction of executed volume of all order volume), and the fill-rate for lit orders on the fraction of volume that is tradedwith dark orders and a number of control variables. In the first stage, the variable of interest, % dark tradingit, is instrumented by the fraction of darkorders in the preceding time period, % dark ordersit−1. Control variables are the daily realization of the Canadian volatility index VIXC, the averageorder size in the stock in the preceding week, and the contemporaneous realizations of the logarithm of the time-weighted average mid-quote (over the10-minute interval), and market fragmentation (measure by one minus the Herfindahl Index of market fragmentation in Canada). The sample consistsof 229 companies that are in the TSX Composite Index and it covers the time span from July 01, 2011 to December 30, 2011. All regressions includestock fixed effects and fixed effects for the 10-minute interval. Standard errors are double-clustered by firm and time. * indicates significance of non-zerocorrelation at the 10% level, *+ at the 6% level, ** at the 5% level, **+ at the 2% level, and *** at the 1% level.
Second stage of the intra-day IV estimationMain instrument: % dark ordersit−1
Dependent variable mid-quote volatility log-$- volume Fill rate all orders Fill rate lit orders
(1) (2) (3) (1) (2) (3) (1) (2) (3) (1) (2) (3)
% dark tradingit -0.3585*** -0.3389*** -0.3513*** 0.0076*** 0.0067*** 0.0075*** 0.0481*** 0.0403*** 0.0488*** 0.0892*** 0.0811*** 0.0899***(0.025) (0.024) (0.024) (0.001) (0.001) (0.001) (0.008) (0.008) (0.008) (0.011) (0.010) (0.011)
VIXCd -0.0643 -0.2367*** -0.2341*** 0.0133*** 0.0162*** 0.0160*** -0.0825*** -0.0970*** -0.0989*** -0.0865*** -0.1015*** -0.1034***(0.047) (0.054) (0.054) (0.002) (0.002) (0.002) (0.009) (0.010) (0.010) (0.009) (0.010) (0.010)
log mid-quoteit -23.2330*** -23.2929*** 0.3653*** 0.3693*** -2.2788*** -2.2375*** -2.3596*** -2.3181***(3.057) (3.085) (0.120) (0.123) (0.594) (0.611) (0.613) (0.630)
Fragmentationit 15.9162*** 15.9363*** -1.0724*** -1.0727*** -10.9687*** -10.9668*** -11.1266*** -11.1246***(1.064) (1.066) (0.055) (0.055) (0.511) (0.511) (0.526) (0.526)
Observations 715,212 715,212 715,212 715,212 715,212 715,212 715,159 715,159 715,159 715,150 715,150 715,150adjusted R-squared 0.006 0.005 0.009 0.076 0.051 0.077 0.039 0.003 0.039 0.025 -0.009 0.025
Figure 1
Trading in Canada: Fraction of Dark and Liquidity in 2011
The left panel plots the percentages of average daily per-company dark volume of all continuous Canadian-trading volume (i.e., excluding dealer crosses)for the sample of TSX Composite securities. The data is also split up by MatchNow, Alpha IntraSpread and TSX dark. The right panel plots the averagedaily share of TSX volume that involves dark orders, split by the TSX 60 and the TSX Completion constituents. Vertical lines indicate the event dates fromthe introduction of dark orders for the TSX60 (April 01, 2011) and the TSX Completion (May 20, 2011). All plots contain the levels (thin, light-colouredlines) and ± 15-day moving averages (thick, bold-colored lines).
02
46
810
% D
ark
Tra
ding
01jan2011 01apr2011 01jul2011 01oct2011 01jan2012January 15 − December 15, 2012
Total Dark Trading MA Total Dark Trading TSX Dark Trading MA TSX Dark Trading MatchNow MA MatchNow IntraSpread MA IntraSpread
Dark Trading TSX Composite
01
23
4%
dar
k
01jan2011 01mar2011 01may2011 01jul2011 01sep2011 01nov2011January 15 − November 15, 2012
TSX60 MA TSX60 TSX Completion MA TSX Completion
Dark TSX Trading TSX60 vs TSX Completion
Figure 2
Quoted Liquidity: Bid-Ask Spreads
The left panel plots the average of daily per-company time-weighted quoted bid ask spreads for TSX60 and TSX Completion securities, where spreads aremeasured in cents. The right panel plots the difference of the bid-ask-spreads between TSX60 and TSX Completion securities. The plots also containsa line that displays the fraction of TSX continuous-market volume that is traded with dark orders (note that the bold line shows non-zero dark tradingbefore the event date, simply because it is a forward- and backward-looking moving average). Vertical lines represent the two event dates: the first, April01, marks the date of the introduction of dark orders for TSX60 securities; the second, May 20, marks the introduction date for all remaining securities.As the left panel shows, spreads for TSX60 stocks following the introduction of dark orders. Moreover, as the right panel shows, spreads also increaserelative to TSX Completion stocks. All plots contain the levels (thin, light-coloured lines) and ± 15-day moving averages (thick, bold-colored lines).
0.5
11.
5%
dar
k
22.
53
3.5
bid−
ask
spre
ad in
cen
ts
01feb2011 01mar2011 01apr2011 01may2011 01jun2011 February 15 − May 31, 2012
TSX60 MA TSX60 TSX Completion MA TSX Completion % TSX dark MA % TSX dark
Time−weighted quoted spread: TSX60 vs TSX Completion
0.5
11.
52
% d
ark
−1
−.5
0di
ffere
nce
in b
id−
ask
spre
ad in
cen
ts
01feb2011 01mar2011 01apr2011 01may2011 01jun2011 February 1 − May 31, 2011
Diff TSX60 − TSX Completion MA Diff TSX60 − TSX Completion
% TSX dark MA % TSX dark
Time−weighted quoted spread: TSX60 minus TSX Completion
Figure 3
Quoted Liquidity: Log-$-Depth at the Best Bid and Ask Prices
The left panel plots the average of daily per-company time-weighted quoted log-$-depth for TSX60 and TSX Completion securities. The right panelplots the difference of the log-depths between TSX60 and TSX Completion securities. The plots also contains a line that displays the fraction of TSXcontinuous-market volume that is traded with dark orders (note that the bold line shows non-zero dark trading before the event date, simply because it isa forward- and backward-looking moving average). Vertical lines represent the two event dates: the first, April 01, marks the date of the introduction ofdark orders for TSX60 securities; the second, May 20, marks the introduction date for all remaining securities. As the left panel shows, depth declines forboth TSX60 and TSX Completion constituents. The right panel displays that the average difference in depths before the event was larger than after theevent, suggesting a decline in depth following the introduction of dark orders. However, it is clearly visible that the decline happened before mid-Marchand that afterwards, depth-differences were stable. We thus conclude that that the introduction of dark orders had no effect on depth. All plots containthe levels (thin, light-coloured lines) and ± 15-day moving averages (thick, bold-colored lines).
0.5
11.
5
0.5
11.
52
01feb2011 01mar2011 01apr2011 01may2011 01jun2011 February 15 − May 31, 2012
TSX60 MA TSX60 TSX Completion MA TSX Completion % TSX dark MA % TSX dark
Log−$−Depth TSX60 vs TSX Completion
0.5
11.
52
% d
ark
.7.8
.91
1.1
diffe
renc
e in
log
$−de
pth
01feb2011 01mar2011 01apr2011 01may2011 01jun2011 February 1 − May 31, 2011
Diff TSX60 − TSX Completion MA Diff TSX60 − TSX Completion
% TSX dark MA % TSX dark
Time−weighted quoted depth: TSX60 minus TSX Completion
Figure 4
Intra-Day Patterns of Volume and Dark Trading
The figure displays a the intra-day trading patterns of TSX volume and the share of dark volume of orders and trades. The horizontal axis marks the 3510-minute trading intervals between 9:45 a.m. and 3:35 p.m. The left panel plots the patterns of average total and dark log-$-volume on the TSX overthe 230 TSX Composite securities for the second half of 2011. The total TSX volume is measured on the left vertical axis, and the TSX dark volume ismeasured on the right vertical axis. Both series display a U-shaped pattern. The right panel plots the average share of dark volume of TSX trading andorder volume. As the pale illustrates, dark volume as a share of total trading and order volume is, loosely, flat throughout the day.
0.1
.2.3
.4D
ark
TS
X V
olum
e
1.8
22.
22.
42.
6
0 10 20 30 4010 minute intervals from 9:45−15:35
All TSX continuous volume TSX dark volume
Volume (in logs)
11.
52
2.5
3pe
rcen
t
0 10 20 30 40Interval
% Dark Trading % Dark Orders
% Dark Trading & Orders
Figure 5
Intra-Day Patterns of Liquidity and Volatility
The figure displays a the intra-day trading patterns of bid-ask-spreads, depth and mid-quote volatility. The horizontal axis marks the 35 10-minute tradingintervals between 9:45 a.m. and 3:35 p.m. The left panel plots the patterns of average the average time-weighted quote spread and the volume-weightedeffective spread over the 230 TSX Composite securities for the second half of 2011. Spreads are in cents and measured on the left vertical axis, depth is thelog-$-volume available for trading at the best bid and offer prices on the TSX and it is measured on the right vertical axis. The spread measures declinethroughout the trading day, depth increases. The right panel plots the average intra-day volatility measure by the highest minus the lowest mid-quote,scaled by the average mid-quote, and the time-weighted depth available at the best bid and offer prices, averaged over the 10 minute time interval. Thevolatility measure is in basis points and it displays a reverse J-shaped pattern, depth is in logarithms of the $-volume available for trading and it isincreasing over the trading day.
22.
53
3.5
44.
5ce
nts
0 10 20 30 4010 minute intervals from 9:45−15:35
Time Weighted Quoted Spreads Volume Weighted Effective Spreads
TSX Bid−Ask Spreads
.2.3
.4.5
.6
3040
5060
70ba
sis
poin
ts
0 10 20 30 4010 minute intervals from 9:45−15:35
Mid−quote Volatility (High minus Low Price) Log $−Depth
TSX Depth and Volatility