You are on page 1of 63

Do Dark Pools Harm Price Discovery?


Haoxiang Zhu
Graduate School of Business, Stanford University
June 5, 2012
Abstract
Dark pools are equity trading systems that do not publicly display orders. Orders in dark
pools are matched within the exchange bid-ask spread without a guarantee of execution.
Informed traders are more likely to cluster on the heavy side of the market and therefore face
a lower execution probability in the dark pool, relative to uninformed traders. Consequently,
exchanges are more attractive to informed traders, whereas dark pools are more attractive
to uninformed traders. Under natural conditions, adding a dark pool alongside an exchange
concentrates price-relevant information into the exchange and improves price discovery. Dark
pools that operate as nondisplayed limit order books are more attractive to informed traders
than dark pools that execute orders at the exchange midpoint.
Keywords: dark pools, price discovery, liquidity, fragmentation, equity market structure
JEL Classifications: G12, G14, G18

First version: November 2010. For helpful comments, I am very grateful to Darrell Duffie, Sal Arnuk, Jonathan
Berk, John Beshears, Bradyn Breon-Drish, Robert Burns, Peter DeMarzo, Thomas George, Steven Grenadier,
Frank Hatheway, Dirk Jenter, Ron Kaniel, Arthur Korteweg, Ilan Kremer, Charles Lee, Han Lee, Ian Martin, Jim
McLoughlin, Albert Menkveld, Stefan Nagel, Francisco P´erez-Gonz´alez, Paul Pfleiderer, Monika Piazzesi, Michael
Ostrovsky, Martin Schneider, Ken Singleton, Jeffrey Smith, Ilya Strebulaev, Ingrid Werner (discussant), Mao Ye
(discussant), Ruiling Zeng, and Jeff Zwiebel, as well as seminar participants at Stanford University, Chicago Booth,
Princeton University, University of Illinois, MIT Sloan, NYU Stern, Wharton, UT Austin McCombs, Berkeley Haas,
UCLA Anderson, Northwestern Kellogg, the Western Finance Association annual meeting, the NBER Market Design
Working Group meeting, and the SFS Finance Cavalcade. All errors are my own. Corresponding address: Stanford
Graduate School of Business, 655 Knight Way, Stanford, CA 94305-7298. E-mail: haoxiang.zhu@stanford.edu.
Paper URL: http://ssrn.com/abstract=1712173.
1
1 Introduction
Dark pools are equity trading systems that do not publicly display orders. Some dark
pools passively match buyers and sellers at exchange prices, such as the midpoint of
the exchange bid and offer. Other dark pools essentially operate as nondisplayed limit
order books that execute orders by price and time priority.
In this paper, I investigate the impact of dark pools on price discovery. Contrary
to misgivings expressed by some regulators and market participants, I find that under
natural conditions, adding a dark pool improves price discovery on the exchange.
According to the Securities and Exchange Commission (SEC; 2010), as of September
2009, 32 dark pools in the United States accounted for 7.9% of total equity trading
volume. As of mid-2011, industry estimates from the Tabb Group, a consultancy, and
Rosenblatt Securities, a broker, attribute about 12% of U.S. equity trading volume to
dark pools. The market shares of dark pools in Europe, Canada, and Asia are smaller
but quickly growing (International Organization of Securities Commissions, 2010).
Dark pools have raised regulatory concerns in that they may harm price discovery.
The European Commission (2010), for example, remarks that “[a]n increased use of
dark pools . . . raise[s] regulatory concerns as it may ultimately affect the quality of the
price discovery mechanism on the ‘lit’ markets.” The International Organization of
Securities Commissions (2011) similarly worries that “the development of dark pools
and use of dark orders could inhibit price discovery if orders that otherwise might
have been publicly displayed become dark.” According to a recent survey conducted
by the CFA Institute (2009), 71% of respondents believe that the operations of dark
pools are “somewhat” or “very” problematic for price discovery. The Securities and
Exchange Commission (2010), too, considers “the effect of undisplayed liquidity on
public price discovery” an important regulatory question. Speaking of nondisplayed
liquidity, SEC Commissioner Elisse Walter commented that “[t]here could be some
truth to the criticism that every share that is crossed in the dark is a share that
doesn’t assist the market in determining an accurate price.”
1
My inquiry into dark pools builds on a simple model of strategic venue selection
by informed and liquidity traders. Informed traders hope to profit from proprietary
information regarding the value of the traded asset, whereas liquidity traders wish to
meet their idiosyncratic liquidity needs. Both types of traders optimally choose between
an exchange and a dark pool. The exchange displays a bid and an ask and executes
all submitted orders at the bid or the ask. The dark pool free-rides on exchange prices
1
“Speech by SEC Commissioner: Opening Remarks Regarding Dark Pools,” October 21, 2009.
2
by matching orders within the exchange’s bid and ask. Unlike the exchange, the dark
pool has no market makers through which to absorb excess order flow and thus cannot
guarantee execution. Sending an order to the dark pool therefore involves a trade-off
between potential price improvement and the risk of no execution.
Execution risk in the dark pool drives my results. Because matching in the dark
pool depends on the availability of counterparties, some orders on the “heavier” side
of the market—the side with more orders—will fail to be executed. These unexecuted
orders may suffer costly delays. Because informed orders are positively correlated with
the value of the asset and therefore with each other, informed orders are more likely to
cluster on the heavy side of the market and suffer lower execution probabilities in the
dark pool. By contrast, liquidity orders are less correlated with each other and less likely
to cluster on the heavy side of the market; thus, liquidity orders have higher execution
probabilities in the dark pool. This difference in execution risk pushes relatively more
informed traders into the exchange and relatively more uninformed traders into the
dark pool. Under natural conditions, this self selection lowers the noisiness of demand
and supply on the exchange and improves price discovery.
The main intuition underlying my results does not hinge on the specific trading
mechanisms used by a dark pool. For example, a dark pool may execute orders at the
midpoint of the exchange bid and ask or operate as a nondisplayed limit order book.
As I show, with both of these mechanisms, traders face a trade-off between potential
price improvement and execution risk. Dark pools that operate as limit order books
are, however, relatively more attractive to informed traders because limit orders can
be used to gain execution priority and thus reduce execution risk. This result suggests
that informed traders have even stronger incentives to trade on the exchange under a
“trade-at” rule, which requires that trading venues that do not quote the best price
either to route incoming orders to venues quoting the best price or to provide incoming
orders with a sufficiently large price improvement over the prevailing best price. The
impact of a trade-at rule on price discovery complements previous fairness-motivated
arguments that displayed orders—which contribute to pre-trade transparency—should
have strictly higher priority than do nondisplayed orders at the same price.
2
Dark pools do not always improve price discovery. For example, in the unlikely event
that liquidity traders push the net order flow far opposite of the informed traders, the
2
For example, the Joint CFTC-SEC Advisory Committee (2011) has noted: “Under current Regulation NMS
routing rules, venues cannot ‘trade through’ a better price displayed on another market. Rather than route the
order to the better price, however, a venue can retain and execute the order by matching the current best price
even if it has not displayed a publicly accessible quote order at that price. While such a routing regime provides
order execution at the current best displayed price, it does so at the expense of the limit order posting a best price
which need not receive execution.”
3
presence of a dark pool can exacerbate the misleading inference regarding the asset
value. Moreover, better price discovery needs not coincide with higher liquidity or
welfare. Indeed, more informative orders often lead to better price discovery but also
tend to worsen adverse selection on the exchange, which results in wider spreads and
higher price impacts. The welfare implications of dark pools could naturally depend
on elements outside the setting of my model, such as how price discovery and liquidity
affect production decisions, asset allocation, and capital formation.
3
In addition, for
analytical tractability I have abstracted from some of the trading practices that are
applied in dark pools, such as “pinging,” order routing, and “indication of interest”
(IOI).
4
These and other procedures used by some dark pools may well contribute to
concerns regarding their impact on price discovery, although these practices are distinct
from the implications of execution risk, which I focus on in this paper. Finally, the price-
discovery effect of dark pools complements their “size discovery” function, by which
large institutional orders are executed without being revealed to the broad market.
This size-discovery benefit of dark trading has been widely acknowledged by market
participants and regulators, and today only a handful of dark pools execute large orders
(Securities and Exchange Commission, 2010; Ready, 2012).
To the best of my knowledge, this paper is the first to show that the addition of
a dark pool can improve price discovery. My finding stands in contrast to that of Ye
(2011), who studies the venue choice of a large informed trader in the Kyle (1985)
framework and concludes that the addition of a dark pool harms price discovery on
the exchange. Ye (2011), however, assumes exogenous choices of trading venues by
liquidity traders, whereas the endogenous venue choices of liquidity traders are critical
to my results. Most other existing models of dark pools either exogenously fix the
strategies of informed traders, as in Hendershott and Mendelson (2000), or do not
consider the role of asymmetric information regarding the asset value, as in Degryse,
Van Achter, and Wuyts (2009) and Buti, Rindi, and Werner (2011b). Going beyond
the midpoint-matching mechanism, my study additionally reveals that dark pools with
more discretion in execution prices are more attractive to informed traders.
The focus of this paper—i.e., on the fragmentation of order flow between an ex-
3
For example, see Bond, Edmans, and Goldstein (2012) for a survey on the literature that studies the effects of
financial markets on the real economy.
4
“Pinging” orders are marketable orders that seek to interact with displayed or nondisplayed liquidity. Pinging
is sometimes used to learn about the presence of large hidden orders. Order routing means sending orders from
venue to venue, typically by algorithms. For example, if a dark pool cannot execute an order because there is no
counterparty, the dark pool can route the order to another dark pool, which may further route the order into the
market. An IOI is an electronic message that contains selected information (such as the ticker) about an order and
is sent by a trading venue (such as a dark pool or a broker) to a selected group of market participants in order to
facilitate a match.
4
change and a dark pool—differs from the focus of prior studies on competition among
multiple markets. In exchange markets, for example, informed traders and liquidity
traders tend to cluster by time (Admati and Pfleiderer, 1988) or by location (Pagano,
1989; Chowdhry and Nanda, 1991). However, as modeled here, informed traders clus-
ter less with liquidity traders in the dark pool than on the exchange because informed
traders face higher execution risk in the dark pool. Related to the effect captured by
my model, Easley, Keifer, and O’Hara (1996) suggest that the purchase of retail order
flows (“cream-skimming”) by regional exchanges results in higher order informativeness
on the NYSE. In contrast with the mechanism studied in their paper, in my model dark
pools rely on self selection, rather than intermediaries, to separate, at least partially,
informed traders from liquidity traders.
My results have several empirical implications. For example, the model predicts
that higher order imbalances tend to cause lower dark pool activity; higher volumes
of dark trading lead to wider spreads and higher price impacts on exchanges; volume
correlation across stocks is higher on exchanges than in dark pools; and informed traders
more actively participate in dark pools when asymmetric information is more severe
or when the dark pool allows more discretion in execution prices. Section 6 discusses
these implications, as well as discussing recent, related empirical evidence documented
by Ready (2012), Buti, Rindi, and Werner (2011a), Ye (2010), Nimalendran and Ray
(2012), Degryse, de Jong, and van Kervel (2011), Jiang, McInish, and Upson (2011),
O’Hara and Ye (2011), and Weaver (2011), among others.
2 An Overview of Dark Pools
This section provides an overview of dark pools. I discuss why dark pools exist, how
they operate, and what distinguishes them from each other. For concreteness, I tailor
this discussion for the market structure and regulatory framework in the United States.
Dark pools in Europe, Canada, and Asia operate similarly.
Before 2005, dark pools had low market share. Early dark pools were primarily
used by institutions to trade large blocks of shares without revealing their intentions
to the broad market, in order to avoid being front-run.
5
A watershed event for the
U.S. equity market was the adoption in 2005 of Regulation National Market System,
or Reg NMS (Securities and Exchange Commission, 2005), which abolished rules that
had protected the manual quotation systems of incumbent exchanges. In doing so,
5
Such predatory trading is modeled by Brunnermeier and Pedersen (2005) and Carlin, Lobo, and Viswanathan
(2007).
5
Figure 1: U.S. equity trading volume and the market share of dark pools. The left axis plots the
daily consolidated equity trading volume in the United States, estimated by Tabb Group. The
right axis plots the market shares of dark pools as a percentage of the total consolidated volume,
estimated by Tabb Group and Rosenblatt Securities.
2009 2010 2011
0
2
4
6
8
10
12
14
16
B
i
l
l
i
o
n

s
h
a
r
e
s


Average daily volume (left)
4
6
8
10
12
14
16
D
a
r
k

p
o
o
l

m
a
r
k
e
t

s
h
a
r
e

(
%
)


Tabb
Rosenblatt
Reg NMS encouraged newer and faster electronic trading centers to compete with the
incumbents. Since Reg NMS came into effect, a wide variety of trading centers have
been established. As of September 2009, the United States had about 10 exchanges,
5 electronic communication networks (ECNs), 32 dark pools, and over 200 broker-
dealers (Securities and Exchange Commission, 2010). Exchanges and ECNs are referred
to as transparent, or “lit,” venues; dark pools and broker-dealer internalization are
considered opaque, or “dark,” venues. In Europe, the adoption in 2007 of the Markets
in Financial Instruments Directive (MiFID) similarly led to increased competition and
a fast expansion of equity trading centers.
6
Figure 1 shows the consolidated volume of U.S. equity markets from July 2008 to
June 2011, as well as the market share of dark pools during the same periods, estimated
by Tabb Group and Rosenblatt Securities. According to their data, the market share
of dark pools roughly doubled from about 6.5% in 2008 to about 12% in 2011, whereas
consolidated equity volume dropped persistently from about 10 billion shares per day
in 2008 to about 7 billion shares per day in 2011. A notable exception to the decline
in consolidated volume occurred around the “Flash Crash” of May 2010.
Dark pools have gained market share for reasons that go beyond recent regulations
6
For example, according to CFA Institute (2009), European equity market had 92 regulated markets (exchanges),
129 “multilateral trading facilities” (MTFs), and 13 “systematic internalizers” as of Septempber 2010. For more
discussion of MiFID and European equity market structure, see European Commission (2010).
6
designed to encourage competition. Certain investors, such as institutions, simply need
nondisplayed venues to trade large blocks of shares without alarming the broad market.
This need has increased in recent years as the order sizes and depths on exchanges
have declined dramatically (Chordia, Roll, and Subrahmanyam, 2011). Further, dark
pools attract investors by offering potential price improvements relative to the best
prevailing bid and offer on exchanges. Finally, broker-dealers handling customer orders
have strong incentives to set up their own dark pools, where they can better match
customer orders internally and therefore save trading fees that would otherwise be paid
to exchanges and other trading centers.
Dark pools differ from each other in many ways. We can categorize them, roughly,
into the three groups shown in Table 1.
Dark pools in the first group match customer orders by acting as agents (as opposed
to trading on their own accounts). In this group, transaction prices are typically derived
from lit venues. These derived prices include the midpoint of the national best bid and
offer (NBBO) and the volume-weighted average price (VWAP). Examples in this group
include block-crossing dark pools such as ITG Posit and Liquidnet.
7
Posit crosses orders
a few times a day at scheduled clock times (up to some randomization), although in
recent years it has also offered continuous crossing. Liquidnet is integrated into the
order-management systems of institutional investors and alerts potential counterparties
when a match is found. Instinet is another agency broker that operates scheduled
and continuous dark pools. Dark pools operated by exchanges typically use midpoint
matching as well. Because Group-1 dark pools rely on lit venues to determine execution
prices, they typically do not provide direct price discovery.
Within the second group, dark pools operate as continuous nondisplayed limit order
books, accepting market, limit, or “pegged” orders.
8
This group includes many of
the dark pools owned by major broker-dealers, including Credit Suisse Crossfinder,
Goldman Sachs Sigma X, Citi Match, Barclays LX, Morgan Stanley MS Pool, and
UBS PIN. Unlike Group-1 dark pools that execute orders at the market midpoint or
VWAP, Group-2 dark pools derive their own execution prices from the limit prices of
submitted orders. Price discovery can therefore take place. Another difference is that
Group-2 dark pools may contain proprietary order flows from the broker-dealers that
operate them. In this sense, these dark pools are not necessarily “agency only.”
Dark pools in the third group act like fast electronic market makers that immediately
accept or reject incoming orders. Examples include Getco and Knight. Like the second
7
See also Ready (2012) for a discussion of these two dark pools.
8
Pegged orders are limit orders with the limit price set relative to an observable market price, such as the bid,
the offer, or the midpoint. As the market moves, the limit price of a pegged order moves accordingly.
7
Table 1: Dark pool classification by trading mechanisms.
Types Examples Typical features
Matching at ex-
change prices
ITG Posit, Liquidnet, In-
stinet
Mostly owned by agency brokers
and exchanges; typically execute or-
ders at midpoint or VWAP, and
customer-to-customer
Nondisplayed
limit order books
Credit Suisse Crossfinder,
Goldman Sachs Sigma X,
Citi Match, Barclays LX,
Morgan Stanley MS Pool,
UBS PIN
Most broker-dealer dark pools; may
offer some price discovery and con-
tain proprietary order flow
Electronic market
makers
Getco and Knight High-speed systems handling
immediate-or-cancel orders; typi-
cally trade as principal
group, transaction prices on these platforms are not necessarily calculated from the
national best bid and offer using a transparent rule. In contrast with dark pools in
Groups 1 and 2, Group-3 dark pools typically trade on their own accounts as principals
(as opposed to agents or marketplaces).
Appendix A discusses additional institutional features of dark pools. Overviews
of dark pools and nondisplayed liquidity are also provided by Johnson (2010), Butler
(2007), Carrie (2008), Securities and Exchange Commission (2010), European Commis-
sion (2010), CSA/IIROC (2009), and International Organization of Securities Commis-
sions (2011).
3 Modeling the Exchange and the Dark Pool
This section presents a two-period model of trading-venue selection. Each trader
chooses whether to trade on a transparent exchange or in a dark pool. The dark pool
modeled in this section passively matches orders at the midpoint of the exchange’s bid
and ask. Section 4 models a dark pool that operates as a nondisplayed limit order
book. The order-book setting provides additional insights regarding the effect of the
dark pool crossing mechanism for price discovery. A dynamic equilibrium with sequen-
tial arrival of traders is characterized in Section 5. A glossary of key model variables
can be found in Appendix D.
8
3.1 Markets and traders
There are two trading periods, denoted by t = 1, 2. At the end of period 2, an asset
pays an uncertain dividend v that is equally likely to be +σ or −σ. Thus, σ > 0 is the
volatility of the asset value. The asset value v is publicly revealed at the beginning of
period 2.
Two trading venues operate in parallel: a lit exchange and a dark pool. The ex-
change is open in periods 1 and 2. On the exchange, a risk-neutral market maker sets
competitive bid and ask prices. Market orders sent to the exchange arrive simultane-
ously. Exchange buy orders are executed at the ask; exchange sell orders are executed
at the bid. The exchange here is thus similar to that modeled by Glosten and Milgrom
(1985).
9
After period-1 orders are executed, the market maker announces the volume
V
b
of exchange buy orders and the volume V
s
of exchange sell orders. The market
maker also announces the exchange closing price P
1
, which is the expected asset value,
conditional on V
b
and V
s
. The closing price P
1
is also the price at which the market
maker is willing to execute a marginal order at the end of period 1. A key objec-
tive of this section is to analyze price discovery, that is, the informativeness of these
announcements, in particular P
1
, for the fundamental value v of the asset.
The dark pool executes (or “crosses”) orders in period 1 and is closed in period 2.
Closing the the dark pool in period 2 is without loss of generality because once the
dividend v is announced in period 2, exchange trading is costless. An order submitted to
the dark pool is not observable to anyone but the order submitter. The execution price
of dark pool trades is the midpoint of the exchange bid and ask, also known simply as
the “midpoint” or “mid-market” price. In the dark pool, orders on the “heavier side”—
the buyers’ side if buy orders exceed sell orders, and the sellers’ side if sell orders exceed
buy orders—are randomly selected for matching with those on the “lighter” side. For
example, if the dark pool receives Q
B
buy orders and Q
S
< Q
B
sell orders, all of the
same size, then Q
S
of the Q
B
buy orders are randomly selected, equally likely, to be
executed against the Q
S
sell orders at the mid-market price. Unmatched orders are
returned to the order submitter at the end of period 1. As described in Section 2, this
midpoint execution method is common in dark pools operated by agency brokers and
exchanges. An alternative dark pool mechanism, a nondisplayed limit order book, is
modeled in Section 4.
For-profit traders and liquidity traders, all risk-neutral, arrive at the beginning of
period 1. There is an infinite set of infinitesimal traders of each type. For-profit traders
9
As I describe shortly, the model of this section is not exactly the same as that of Glosten and Milgrom (1985)
because orders here arrive in batches, instead of sequentially. Sequential arrival of orders is considered in Section 5.
9
have a mass of ¯ µ > 0, and can potentially trade one unit of the asset per capita.
10
For-
profit traders can acquire, at a cost, perfect information about v, and thus become
informed traders. These information-acquisition costs are distributed across for-profit
traders, with the cumulative distribution function F : [0, ∞) →[0, 1]. After observing
v, informed traders submit buy orders (in either venue) if v = +σ and submit sell
orders if v = −σ. For-profit traders who do not acquire the information do not trade.
I let µ
I
be the mass of informed traders; their signed trading interest is therefore
Y = sign(v) · µ
I
.
Liquidity buyers and liquidity sellers arrive at the market separately (not as netted).
The mass Z
+
of liquidity buy orders and the mass Z

of liquidity sell orders are
non-negative, independent and identically distributed on [0, ∞) with positive density
functions, and infinitely divisible. Infinite divisibility means that, for each integer
n, the total liquidity buy orders Z
+
can be viewed as the aggregate demand by n
liquidity buyers, whose order sizes are independently and identically distributed random
variables. A similar construction applies for the total liquidity sell orders Z

. Thus,
we can interpret a market with infinitely many liquidity traders as the “limiting case”
of a market with n liquidity buyers and n liquidity sellers as n → ∞.
11
In particular,
because, in the limit, each liquidity trader’s order size has zero mean and zero variance,
the conditional joint distribution of Z
+
and Z

, given this liquidity trader’s order size,
is the same as the unconditional joint distribution of Z
+
and Z

.
12
I denote by 0.5µ
z
10
Trading one unit per capita is without loss of generality because each informed trader is infinitesimal and has
zero mass. As long as per-capita trading size is finite, an informed trader’s order still has zero mass, and the
qualitative nature of equilibria does not change.
11
More specifically, for each integer n, Z
+
can be constructed as the sum of n independently and identically
distributed random variables {Z
+
in
}. That is,
Z
+

n

i=1
Z
+
in
.
Note that the distribution of Z
+
in
depends on n. I assume that the variance of Z
+
in
is finite. Similarly, there exist n
i.i.d. random variables Z

in
such that
Z


n

i=1
Z

in
.
In this setting, {Z
+
in
} and {Z

in
} can be viewed as the order sizes of n liquidity buyers and n liquidity sellers. As
n →∞, the mean and variance of Z
+
in
and Z

in
converge to zero, and liquidity buyers and sellers become infinitesimal.
12
We denote by Φ the probability distribution of Z
+
and show that, for each i, the conditional distribution of

n
j=1
Z
+
jn
, given Z
+
in
, converges to the prior distribution of Z
+
as n →∞. That is, for all z > 0, Φ(z | Z
+
in
) →Φ(z)
as n → ∞. By the independence of Z
+
in
and {Z
+
jn
}
j=i
, this amounts to showing that Z
+
in
converges to zero in
distribution. Indeed, for any z > 0, using Markov’s inequality and the fact that E(Z
+
in
) converges to zero as n →∞,
we have
P(Z
+
in
< z) = 1 −P(Z
+
in
≥ z) ≥ 1 −
E(Z
+
in
)
z
→1, as n →∞.
Similarly, the conditional distribution of Z

, given Z
+
in
, converges in n to the prior distribution of Z

. The proof
10
Figure 2: Time line of the two-period model.
Period 1 Period 2 | |
Traders select venue
or delay trade
Orders executed
Exchange announces
closing price
Value announced v
Remaining orders
executed on exchange
Value paid v
the mean of Z
+
(and Z

) and by 0.5σ
2
z
the variance of Z
+
(and Z

).
Liquidity traders must hold collateral to support their undesired risky positions.
For each liquidity trader, the minimum collateral requirement per unit asset held is the
expected loss, conditional on a loss, of her undesired position. For example, a liquidity
buyer who is already short one unit of the asset has a loss of σ if v = σ, and a gain
of σ if v = −σ. The collateral requirement in this case is σ. For trader i, each unit of
collateral has a funding cost of γ
i
per period. A delay in trade is therefore costly. These
funding costs {γ
i
} are distributed across liquidity traders, with a twice-differentiable
cumulative distribution function G : [0, Γ) → [0, 1], for some Γ ∈ (1, ∞]. Failing to
trade in period 1, liquidity buyer i thus incurs a delay cost of
c
i
= γ
i
E[max(v, 0) | v > 0] = γ
i
σ (1)
per unit of undesirable asset position. A like delay cost applies to liquidity sellers. We
could alternatively interpret this delay cost as stemming from risk aversion or illiquidity.
The key is that liquidity traders differ in their desires for immediacy, captured by the
delay cost c
i
= γ
i
σ. The delay costs of informed traders, by contrast, stem from the
loss of profitable trading opportunities after v is revealed in period 2.
Finally, random variables v, Z
+
, Z

, and the costs of information-acquisition and
delay are all independent, and their probability distributions are common knowledge.
Realizations of Y , Z
+
and Z

are unobservable, with the exception that informed
traders observe v, and hence know Y . Informed and liquidity traders cannot post limit
orders on the exchange; they can trade only with the exchange market maker or by
sending orders to the dark pool.
Figure 2 illustrates the sequence of actions in the two-period model.
for a liquidity seller’s inference is symmetric.
11
3.2 Equilibrium
An equilibrium consists of the quoting strategy of the exchange market maker, the mar-
ket participation strategies of for-profit traders, and the trading strategies of informed
and liquidity traders. In equilibrium, the competitive market maker breaks even in
expectation, and all traders maximize their expected net profits.
Specifically, I let α
e
and α
d
be candidates for the equilibrium fractions of liquidity
traders who, in period 1, send orders to the exchange and to the dark pool, respectively.
The remainder, α
0
= 1 − α
e
− α
d
, choose not to submit orders in period 1 and delay
trade to period 2. We let β be the period-1 fraction of informed traders who send
orders to the dark pool. The remaining fraction 1 − β of informed traders trade on
the exchange. (Obviously, informed traders never delay their trades as they will have
lost their informational advantage by period 2.) Once the asset value v is revealed in
period 2, all traders who have not traded in period 1—including those who deferred
trading and those who failed to execute their orders in the dark pool—trade with the
market maker at the unique period-2 equilibrium price of v.
I first derive the equilibrium exchange bid and ask, assuming equilibrium partici-
pation fractions (β, α
d
, α
e
). Because of symmetry and the fact that the unconditional
mean of v is zero, the midpoint of the market maker’s bid and ask is zero. There-
fore, the exchange ask is some S > 0, and the exchange bid is −S, where S is the
exchange’s effective spread, the absolute difference between the exchange transaction
price and the midpoint. For simplicity, I refer to S as the “exchange spread.” As in
Glosten and Milgrom (1985), the exchange bid and ask are set before exchange orders
arrive. Given the participation fractions (β, α
d
, α
e
), the mass of informed traders on
the exchange is (1 −β)µ
I
, and the expected mass of liquidity traders on the exchange
is α
e
E(Z
+
+ Z

) = α
e
µ
z
. Because the market maker breaks even in expectation, we
have that
0 = −(1 −β)µ
I
(σ −S) +α
e
µ
z
S, (2)
which implies that
S =
(1 −β)µ
I
(1 −β)µ
I

e
µ
z
σ. (3)
The dark pool crosses orders at the mid-market price of zero.
Next, I derive the equilibrium mass µ
I
of informed traders. Given the value σ of
information and the exchange spread S, the net profit of an informed trader is σ −S.
The information-acquisition cost of the marginal for-profit trader, who is indifferent
12
between paying for information or not, is also σ−S. Because all for-profit traders with
lower information-acquisition costs become informed, the mass of informed traders in
equilibrium is ¯ µF(σ−S), by the exact law of large numbers (Sun, 2006). We thus have
µ
I
= ¯ µF (σ −S) = ¯ µF
_
α
e
µ
z
(1 −β)µ
I

e
µ
z
σ
_
. (4)
For any fixed β ≥ 0 and α
e
> 0, (4) has a unique solution µ
I
∈ (0, ¯ µ).
Finally, I turn to the equilibrium trading strategies. Without loss of generality, I
focus on the strategies of buyers. In the main solution step, I calculate the expected
payoffs of an informed buyer and a liquidity buyer, on the exchange and in the dark pool.
The equilibrium is then naturally determined by conditions characterizing marginal
traders who are indifferent between trading on the exchange and in the dark pool.
Suppose that α
d
> 0. Because informed buyers trade in the same direction, they
have the dark pool crossing probability of
r

= E
_
min
_
1,
α
d
Z

α
d
Z
+
+βµ
I
__
, (5)
where the denominator and the numerator in the fraction above are the masses of
buyers and sellers in the dark pool, respectively. Liquidity buyers, on the other hand,
do not observe v. If informed traders are buyers, then liquidity buyers have the crossing
probability r

in the dark pool. If, however, informed traders are sellers, then liquidity
buyers have the crossing probability
r
+
= E
_
min
_
1,
α
d
Z

+βµ
I
α
d
Z
+
__
. (6)
Obviously, for all β > 0, we have
1 > r
+
> r

> 0. (7)
Because liquidity buyers assign equal probabilities to the two events {v = +σ} and
{v = −σ}, their dark pool crossing probability (r
+
+ r

)/2 is greater than informed
traders’ crossing probability r

. In other words, correlated informed orders have a lower
execution probability in the dark pool than relatively uncorrelated liquidity orders.
If the dark pool contains only liquidity orders (that is, β = 0), then any dark pool
13
buy order has the execution probability
¯ r = E
_
min
_
1,
Z

Z
+
__
. (8)
For our purposes, ¯ r measures the degree to which liquidity orders are balanced. Per-
fectly balanced liquidity orders correspond to ¯ r = 1. For α
d
= 0, I define r
+
= r

= 0.
The expected profits of an informed buyer on the exchange and in the dark pool
are, respectively,
W
e
= σ −S, (9)
W
d
= r

σ. (10)
I denote by c the delay cost of a generic liquidity buyer per unit of asset position.
This buyer’s per-unit net payoffs of deferring trade, trading on the exchange, and
trading in the dark pool are, respectively,
X
0
(c) = −c, (11)
X
e
= −S, (12)
X
d
(c) = −
r
+
−r

2
σ −c
_
1 −
r
+
+r

2
_
. (13)
The terms on the right-hand side of (13) are the liquidity trader’s adverse selection
cost and delay cost in the dark pool, respectively. For β > 0, crossing in the dark pool
implies a positive adverse selection cost because execution is more likely if a liquidity
trader is on the side of the market opposite to that of informed traders. For β = 0,
this adverse-selection cost is zero. For simplicity, in the remaining of the paper the net
profits and delay costs of liquidity traders refer to profits and costs per unit of asset,
unless otherwise specified. It is without loss of generality to focus on the venue decision
for one unit of asset because, by risk neutrality, each trader’s optimal venue choice is
a corner solution with probability one.
From (9) and (12), W
e
−X
e
= σ. For all delay cost c ≤ σ,
W
d
−X
d
(c) =
r
+
+r

2
σ +c
_
1 −
r
+
+r

2
_
≤ σ = W
e
−X
e
. (14)
That is, provided c ≤ σ, the dark pool is more attractive to liquidity traders than to
informed traders, relative to the exchange. In particular, (14) implies that a liquidity
trader with a delay cost of σ (or a funding cost of γ = 1) behaves in the same way as
14
an informed trader. In addition,
X
d
(c) −X
0
(c) = −
r
+
−r

2
σ +
r
+
+r

2
c. (15)
So a liquidity trader with a funding cost of γ = (r
+
− r

)/(r
+
+ r

) is indifferent
between deferring trade and trading in the dark pool.
Proposition 1. There exists a unique threshold volatility ¯ σ > 0 such that:
1. If σ ≤ ¯ σ, then there exists an equilibrium (β = 0, α
d
= α

d
, α
e
= 1 − α

d
), where
α

d
∈ (0, G(1)] and µ

I
solve
G
−1

d
)(1 − ¯ r) =
µ
I
µ
I
+ (1 −α
d

z
, (16)
µ
I
= ¯ µF
_
(1 −α
d

z
µ
I
+ (1 −α
d

z
σ
_
. (17)
2. If and only if σ > ¯ σ, there exists an equilibrium (β = β

, α
d
= α

d
, α
e
= 1 −G(1)),
where β

, α

d
∈ (0, G(1)], and µ

I
solve
r

= 1 −
(1 −β)µ
I
(1 −β)µ
I
+ (1 −G(1))µ
z
, (18)
α
d
= G(1) −G
_
r
+
−r

r
+
+r

_
, (19)
µ
I
= ¯ µF
_
(1 −G(1))µ
z
(1 −β)µ
I
+ (1 −G(1))µ
z
σ
_
. (20)
The proof of Proposition 1 is provided in Appendix C, but we outline its main
intuition here. If the volatility σ is sufficiently low, the exchange spread is low; thus,
the price-improvement benefit of the dark pool is lower than the cost of execution risk.
In this case, informed traders avoid the dark pool (i.e. β = 0). The equilibrium is then
determined by the marginal liquidity trader who is indifferent between trading on the
exchange and trading in the dark pool, as well as by the marginal for-profit trader who
is indifferent about whether to acquire the information.
If the volatility σ is sufficiently high, informed traders joint liquidity traders in
the dark pool to avoid the higher exchange spread. Thus, β ∈ (0, 1). In this case,
the equilibrium is determined by three indifference conditions. First, informed traders
must be indifferent between trading in either venue, as shown in (18). By (14), a
liquidity trader with a delay cost of σ is also indifferent between the two venues. Thus,
α
0

d
= G(1) and α
e
= 1−G(1). The second indifference condition (19) then follows
15
from (15). Here, the fraction α
0
of liquidity traders who delay trade must be strictly
positive because informed traders introduce adverse selection into the dark pool. The
third condition (20) says that the marginal for-profit trader is indifferent about whether
to acquire the information.
Similarly, we can characterize an equilibrium for a market structure in which only
the exchange is operating and the dark pool is absent. This exchange-only equilibrium,
stated below, may also be interpreted as one in which a dark pool is open but no trader
uses it.
Corollary 1. With only an exchange and no dark pool, there exists an equilibrium in
which β

= α

d
= 0, and µ

I
and α

e
∈ (1 −G(1), 1) solve
µ
I
µ
I

e
µ
z
= G
−1
(1 −α
e
) (21)
µ
I
= ¯ µF
_
α
e
µ
z
µ
I

e
µ
z
σ
_
. (22)
Equilibrium selection
The equilibria characterized in Proposition 1 need not be unique among all equilibria
solving (16)-(17) and (18)-(20). For example, under the condition (63), both sides of
(16) strictly increase in α
d
. Similarly, both sides of (19) strictly increase in α
d
, and
both sides of (21) strictly decrease in α
e
. Thus, given the absence of a single-crossing
property, multiple equilibria may arise.
13
I use stability as an equilibrium selection criterion, which allows me to compute the
comparative statics of the selected equilibria. Among the equilibria characterized by
Case 1 of Proposition 1, I select that with the smallest liquidity participation α

d
in the
dark pool among those with the property that, as α
d
varies in the neighborhood of α

d
,
the left-hand side of (16) crosses the right-hand side from below.
14
Under the condi-
tions of Proposition 1, this equilibrium exists and is robust to small perturbations.
15
Moreover, once α
d
is determined in equilibrium, µ
I
and β are uniquely determined,
too, as shown in the proof of Proposition 1.
13
One special condition that guarantees the uniqueness of the equilibrium in Case 1 of Proposition 1 is that the
distribution function G of delay costs is linear. With a linear G, the condition (63) is also necessary for the existence
of solutions to (16)-(17).
14
Selecting the stable equilibrium corresponding to the smallest α

d
is arbitrary but without loss of generality. As
long as the selected equilibrium is stable, comparative statics calculated later follow through.
15
If, for example, α

d
is perturbed to α

d
+ for sufficiently small > 0, then the marginal liquidity trader has a
higher cost in the dark pool than on the exchange, and therefore migrates out of the dark pool. Thus, α
d
is “pushed
back” to α

d
and the equilibrium is restored. There is a symmetric argument for a small downward perturbation to
α

d
−. By contrast, if there is an equilibrium in which, as α
d
varies, the left-hand side of (16) crosses the right-hand
side from above, this equilibrium would not be stable to local perturbations.
16
Similarly, among equilibria characterized by Case 2 of Proposition 1, I select the
one with the smallest liquidity participation α

d
in the dark pool among those with the
property that, as α
d
varies in the neighborhood of α

d
, the left-hand side of (19) crosses
the right-hand side from below. In a market without a dark pool (Corollary 1), I select
the equilibrium with the largest liquidity participation α

e
on the exchange among those
with the property that, as α
e
varies in the neighborhood of α

e
, the left-hand side of
(21) crosses the right-hand side from below. By the argument given for Case 1 of
Proposition 1, these selected equilibria exist and are stable.
3.3 Market characteristics and comparative statics
I now investigate properties of the equilibria characterized by Proposition 1. Proposi-
tion 2 and Proposition 3 below aim to answer two questions:
1. In a market with a dark pool and an exchange, how do market characteristics vary
with the value σ of private information?
2. Given a fixed value σ of private information, how does adding a dark pool affect
market behavior?
Proposition 2. In the equilibrium of Proposition 1:
1. For σ ≤ ¯ σ, the dark pool participation rate α
d
of liquidity traders, the total mass
µ
I
of informed traders, and the scaled exchange spread S/σ are strictly increasing
in σ. The exchange participation rate α
e
= 1 − α
d
of liquidity traders is strictly
decreasing in σ. Moreover, α
d
, µ
I
, and S are continuous and differentiable in σ.
2. For σ > ¯ σ, all of µ
I
, βµ
I
, r
+
, and S/σ are strictly increasing in σ, whereas α
d
and
r

are strictly decreasing in σ. Moreover, β, α
d
, µ
I
, S, r
+
, and r

are continuous
and differentiable in σ.
In the equilibrium of Corollary 1, µ
I
and S/σ are strictly increasing in σ, whereas α
e
is strictly decreasing in σ. Moreover, α
e
, µ
I
, and S are continuous and differentiable
in σ.
Proof. See Appendix C.
Proposition 3. In the equilibria of Proposition 1 and Corollary 1:
1. For σ ≤ ¯ σ, adding a dark pool strictly reduces the exchange participation rate α
e
of liquidity traders and the total mass µ
I
of informed traders. Adding a dark pool
strictly increases the exchange spread S and the total participation rate α
e

d
of
liquidity traders in either venue.
17
2. For σ > ¯ σ, adding a dark pool strictly reduces α
e
. Moreover, adding a dark
pool strictly increases the exchange spread S if and only if, in the equilibrium of
Proposition 1,
r

< 1 −
µ
I
µ
I
+ (1 −G(1 −r

))µ
z
. (23)
It is sufficient (but not necessary) for (23) that
G

(γ) ≤ 0 for all 1 − ¯ r ≤ γ ≤ 1 and F(c) →1 for all c > 0, (24)
Proof. See Appendix C.
We now discuss the intuition and implications of Proposition 2 and Proposition 3
through numerical examples.
3.3.1 Participation rates and exchange spread
The left-hand side plot of Figure 3 shows the equilibrium participation rates in the
exchange and the dark pool. For a small value of information, specifically if σ ≤ ¯ σ,
informed traders trade exclusively on the exchange because the exchange spread is
smaller than the cost of execution risk in the dark pool. An increase in σ widens the
exchange spread, encouraging more liquidity traders to migrate into the dark pool. For
σ > ¯ σ, informed traders use both venues. We observe that informed dark pool partic-
ipation rate β first increases in volatility σ and then decreases. The intuition for this
non-monotonicity is as follows. Consider an increase in the value of information from σ
to σ +, for some > 0. This higher value of information attracts additional informed
traders. For a low β, the dark pool execution risk stays relatively low, and these ad-
ditional informed traders prefer to trade in the dark pool, raising β. For sufficiently
high β, however, informed orders cluster on one side of the dark pool and significantly
reduce their execution probability. Thus, these additional informed traders send orders
to the exchange, reducing β. Nonetheless, the total quantity βµ
I
of informed traders in
the dark pool is strictly increasing in σ. Finally, because informed participation in the
dark pool introduces adverse selection, liquidity traders with low delay costs migrate
out of the dark pool, leading to a decline in their dark pool participation rate α
d
.
The right-hand side plot of Figure 3 shows the scaled exchange spread S/σ. Because
a higher value σ of information encourages more for-profit traders to become informed,
the scaled exchange spread S/σ increases in σ, whether a dark pool is present or not.
For σ ≤ ¯ σ, adding a dark pool raises S/σ by diverting some liquidity traders, but
none of the informed traders, off the exchange. For σ > ¯ σ, adding a dark pool in
18
Figure 3: Participation rates and exchange spread. The left-hand side plot shows the equilibrium
participation rates (β, α
d
, α
e
) in a market with a dark pool. The right-hand side plot shows the
scaled exchange spread S/σ. In both plots, the vertical dotted line indicates the threshold volatility
¯ σ at which the equilibrium of Proposition 1 changes from Case 1 to Case 2. Model parameters:
µ
z
= 60, σ
z
=

60, ¯ µ = 20, Z
+
and Z

have Gamma(30, 1) distributions, G(s) = s/2 for s ∈ [0, 2],
and F(s) = 1 −e
−s/2
for s ∈ [0, ∞).
−2 0 2 4
0
0.2
0.4
0.6
0.8
1
P
a
r
t
i
c
i
p
a
t
i
o
n

r
a
t
e
s


log(σ)
β
α
d
α
e
−2 0 2 4
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
S
/
σ


log(σ)
Lit only
Lit + Dark
this example also increases the scaled spread S/σ because the dark pool diverts more
liquidity traders than informed traders.
3.3.2 Price discovery
Now I turn to price discovery, by which I mean the extent to which the period-1
announcements (P
1
, V
b
, V
s
) are informative of the fundamental asset value v. Since the
market maker observes the volume (V
b
, V
s
), the closing price P
1
is
P
1
= E[v | V
b
, V
s
]. (25)
Because v is binomially distributed, its conditional distribution after period-1 trading
is completely determined by its conditional expectation
E[v | P
1
, V
b
, V
s
] = E[E[v | V
b
, V
s
] | P
1
] = P
1
. (26)
That is, all period-1 public information that is relevant for the asset value v is conveyed
by the closing price P
1
. As we will make precise shortly, the “closer” is P
1
to v, the
better is price discovery.
19
Clearly, P
1
is uniquely determined by the log likelihood ratio
R
1
= log
P(v = +σ | V
b
, V
s
)
P(v = −σ | V
b
, V
s
)
= log
φ
_
Z
+
=
1
αe
[V
b
−(1 −β)µ
I
]
_
· φ
_
Z

=
1
αe
V
s
_
φ
_
Z
+
=
1
αe
V
b
_
· φ
_
Z

=
1
αe
[V
s
−(1 −β)µ
I
]
_,
(27)
where φ is the probability density function of Z
+
and Z

. We have also used the fact
that the prior distribution P(v = +σ) = P(v = −σ) = 0.5.
Given R
1
, the market maker sets the period-1 closing price
P
1
=
e
R
1
−1
e
R
1
+ 1
σ. (28)
Conditional on P
1
, a non-trader assigns the probability
Q
1
≡ P(v = +σ | V
b
, V
s
) =
e
R
1
e
R
1
+ 1
=
1
2
_
P
1
σ
+ 1
_
(29)
that the asset value is high.
Without loss of generality, I condition on v = +σ and consider price discovery to be
unambiguously “improved” if the probability distribution of R
1
is “increased,” in the
sense of first-order stochastic dominance. Complete revelation of v = +σ corresponds
to R
1
= ∞ almost surely.
In general, we need to know the functional form of the density φ(·) in order to
explicitly calculate R
1
, P
1
, and Q
1
. However, since the distribution of Z
+
and Z

is
infinitely divisible, Z
+
and Z

can be expressed as the sums of i.i.d. random variables.
Further, we can always take an example in which, by the central limit theorem, the
density φ(·) is approximated by Normal(0.5µ
z
, 0.5σ
2
z
) when µ
z
and σ
2
z
are sufficiently
large.
16
Substituting into (27) the normal density function, we can approximate R
1
by
R
normal
1
=
2(1 −β)µ
I
α
2
e
σ
2
z
(V
b
−V
s
), (30)
which is the counterpart of R
1
under the normal distribution.
17
Given v = +σ, V
b
−V
s
has a distribution close to that of Normal((1 −β)µ
I
, α
2
e
σ
2
z
), so R
1
is has a distribution
16
We can show this approximation as follows. Fix a small δ > 0 such that m = µ
z
/δ is an integer. By infinite
divisibility, Z
+
can be represented as the sum

m
i=1
Z
+
im
, where {Z
+
im
} are i.i.d. random variables with mean δ and
variance δσ
2
z

z
. Fixing δ, the central limit theorem implies that the distribution of Z
+
is approximately normal
when m is large, that is, when µ
z
and σ
2
z
are large.
17
In the calculation of (30), I have used the central limit theorem and the fact that φ(·) and the normal density
are positive and continuous in [0, ∞).
20
close to that of
Normal
_
2
_
(1 −β)µ
I
α
e
σ
z
_
2
, 4
_
(1 −β)µ
I
α
e
σ
z
_
2
_
∼ Normal
_
2I(β, α
e
)
2
, 4I(β, α
e
)
2
_
,
(31)
where
I(β, α
e
) ≡
(1 −β)µ
I
α
e
σ
z
(32)
is the “signal-to-noise” ratio, which is the mass of informed orders on the exchange
(“signal”) divided by the standard deviation of the imbalance of liquidity orders on
the exchange (“noise”). Naturally, I(β, α
e
) is increasing in the scaled exchange spread
S/σ.
Figure 4 plots the distribution function of R
1
, under normal approximation, with
and without a dark pool. The value σ of information is set to be the threshold value ¯ σ,
so that β = 0 in the equilibria with a dark pool as well as the equilibria without a dark
pool. By Proposition 3, adding a dark pool strictly increases the scaled spread S/σ and
hence the signal-to-noise ratio I(β, α
e
). With a dark pool, the conditional distribution
of R
1
has a higher mean, but also a higher variance. For most realizations of R
1
, and
on average, adding a dark pool decreases the cumulative distribution of R
1
and leads
to a more precise inference of v. Nonetheless, adding a dark pool may increase the
cumulative distribution of R
1
, thus harming price discovery, when the realization of R
1
is sufficiently low.
The price-discovery effect of the dark pool is further illustrated in Figure 5. The
left-hand plot of Figure 5 shows the probability density function of Q
1
, under normal
approximation, with and without a dark pool. As in Figure 4, adding a dark pool
shifts the probability density function of Q
1
to the right, improving price discovery
on average.
18
Nonetheless, the dark pool increases the probability of extremely low
realizations of Q
1
, harming price discovery in these unlikely events. The right-hand
plot of Figure 5 shows how Q
1
depends on the imbalance Z = Z
+
−Z

of liquidity order
flow. Again, for most realizations of Z, adding the dark pool increases Q
1
, improving
price discovery. For unlikely low realizations of Z, adding the dark pool reduces Q
1
,
thus harming price discovery. That is, when the trading interests of liquidity traders
are sufficiently large and opposite in direction to the informed, adding the dark pool
can exacerbate the “misleading” inference regarding the asset value. Because liquidity
trading interests are balanced in expectation, such misleading events are rare, and the
18
We can analytically show that the expectation E[Q
1
] under normal approximation is increasing in the signal-
to-noise ratio I(β, α
e
).
21
Figure 4: Distribution functions of R
1
, under normal approximation, with and without a dark
pool. The true dividend is the threshold value +¯ σ and other parameters are those of Figure 3.
−5 0 5
0.1
0.2
0.3
0.4
P
r
o
b
a
b
i
l
i
t
y

d
e
n
s
i
t
y


R
1
Lit only
Lit + Dark
−5 0 5
0.2
0.4
0.6
0.8
C
u
m
u
l
a
t
i
v
e

d
i
s
t
r
i
b
u
t
i
o
n
R
1
Figure 5: The left-hand plot shows the probability density function of Q
1
, under normal approx-
imation, with and without a dark pool. The right-hand plot shows how Q
1
depends on the order
imbalance Z = Z
+
−Z

of liquidity order flow. Model parameters are those of Figure 4.
0.2 0.4 0.6 0.8
0.5
1
1.5
2
2.5
P
r
o
b
a
b
i
l
i
t
y

d
e
n
s
i
t
y


Q
1
Lit only
Lit + Dark
−40 −20 0 20 40
0
0.2
0.4
0.6
0.8
1
Q
1


Z
Lit only
Lit + Dark
dark pool is normally beneficial for price discovery.
We observe that the practical interpretation of price discovery depends largely on
the horizon of information. Because trading is frequent and fast (with the exception
that large orders can take days to fill), dark pools are most likely to concentrate short-
term information, rather than long-term information, onto the exchange. Short-term
information can be fundamental (such as merger announcements, earnings reports,
or macroeconomic news) or technical (such as the order flows of large institutions).
Moreover, when both short-term investors and long-term investors are present, it is
natural to interpret the former as informed and the latter as uninformed. Under this
22
interpretation, the results of this paper suggest that dark pools are more attractive to
long-term investors than to short-term investors, relative to the exchange.
3.3.3 Dark pool market share
I now calculate the dark pool market share, i.e. the proportion of trading volume
handled by the dark pool. The market share of the dark pool is a direct empirical
measure of dark pool activity. I assume that once the dividend v is announced in
period 2, informed traders who have not yet traded leave the market, because they will
not be able to trade profitably. When calculating the exchange volume, I also include
the transactions of liquidity traders in period 2. Thus, the expected trading volumes
in the dark pool, on the exchange, and in both venues are, respectively,
V
d
= βµ
I
r


d
µ
z
r
+
+r

2
, (33)
V
e
= (1 −β)µ
I

e
µ
z

d
µ
z
_
1 −
r
+
+r

2
_

0
µ
z
, (34)
V = V
e
+V
d
= µ
z

I
(1 −β +βr

). (35)
By Proposition 2, these volumes are differentiable in the volatility σ in each of the two
intervals [0, ¯ σ] and (¯ σ, ∞).
For σ ≤ ¯ σ, the dark pool volume, V
d
= α
d
µ
z
¯ r, is increasing in the volatility σ, by
Proposition 2. In particular, as σ →0, the dark pool participation rate α
d
of liquidity
traders and the dark pool market share V
d
/V converge to zero. For a sufficiently small
σ < ¯ σ, therefore,
d(V
d
/V )

=
d

_
α
d
µ
z
¯ r
µ
z

I
_
=
µ
z
¯ r
µ
z

I
·

d


α
d
µ
z
¯ r

z

I
)
2
·

I

> 0,
where the inequality follows from the fact that lim
σ→0

d
/dσ > 0 (shown in the proof
of Proposition 2). That is, if the volatility σ is sufficiently low, then the dark pool
market share V
d
/V is increasing in σ. For σ ≤ ¯ σ, because the total volume V = µ
z

I
is increasing in σ, the dark pool market share is increasing in the total volume, as
illustrated in the left-hand plot of Figure 6.
Figure 6 further suggests that, as the volatility σ increases beyond ¯ σ, the exchange
volume V
e
can increase substantially, but the dark pool volume V
d
may only increase
mildly or even decline. Thus, the dark pool market share can decrease in volatility
σ for sufficiently large σ, creating a hump-shaped relation between volatility and the
dark pool market share. The model also generates a similar relation between the scaled
23
Figure 6: Expected trading volume on the exchange and in the dark pool. The left-hand plot
shows the volume in the two venues and the market share of the dark pool. The right-hand plot
shows the dark pool market share as a function of the scaled spread S/σ. The vertical dotted line
corresponds to the threshold volatility ¯ σ. Parameters are those of Figure 3.
−2 0 2 4
0
0.1
0.2
0.3
0.4


Dark/Total (left)
−2 0 2 4
0
10
20
30
40
50
60


log(σ)
Lit volume
Dark volume
0.05 0.1 0.15 0.2 0.25 0.3
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4


S/σ
Dark pool market share
spread S/σ and the dark pool market share V
d
/V , as shown on the right-hand plot of
Figure 6.
4 Dark Pools as Nondisplayed Limit Order Books
So far we have studied a dark pool that crosses orders at the midpoint of the exchange
bid and ask. In this section, I model a dark pool that operates as a nondisplayed limit
order book, where execution prices depend on submitted limit orders, as described in
Section 2. Aside from confirming the basic intuition of Section 3, this section offers
additional insights regarding the impact of dark pool mechanisms on the participation
incentives of informed traders.
Although limit-order dark pools may execute orders at prices other than the mid-
point, such price discretion is often limited by “best-execution” regulations. In the
United States, the Order Protection Rule, also known as the “trade through” rule,
stipulates that transaction prices in any market center—including dark pools, ECN,
and broker-dealer internalization—cannot be strictly worse than the prevailing national
best bid and offer (NBBO).
19
For example, if the current best bid is $10 and the best
ask is $10.50, then the transaction price in any market center must be in the interval
[$10, $10.50]. More recently, regulators have also proposed a stricter “trade-at” rule.
19
In Europe, MiFID uses a decentralized best-execution rule, by which investment firms decide whether an
execution works for the best interest of investors.
24
Under a trade-at rule, execution prices in dark pools must be strictly better than the
best bid or offer on all displayed venues, including exchanges. For example, the Joint
CFTC-SEC Advisory Committee (2011) recommends that the SEC consider “its rule
proposal requiring that internalized or preferenced orders only be executed at a price
materially superior (e.g. 50 mils [0.5 cent] for most securities) to the quoted best bid
or offer.”
I now describe and solve a simple model of a limit-order dark pool that operates
under a trade-at rule. The dark pool executes orders by price priority, and I model
its trading mechanism as a double auction. The dark-pool execution price, p

, is
determined such that the aggregate limit buy orders (i.e. demand) at p

is equal to
the aggregate sell limit orders (i.e. supply) at p

. Moreover, I model the effect of a
trade-at rule by assuming that transaction prices in the dark pool must be within the
interval [−xS, xS], where S > 0 is the exchange spread and x ∈ [0, 1] captures the
strictness of the trade-at rule. The trade-through rule currently applied in the United
States corresponds to x = 1, indicating a mandatory price improvement of zero. A
midpoint-matching mechanism corresponds to x = 0, indicating a price improvement
of the entire effective spread S. With the exception of this trade-at rule, the model
of this section is identical to that of Section 3. Proposition 4 below characterizes an
equilibrium that is analogous to Case 1 of Proposition 1.
20
This result sheds light on
how the trade-at rule affects the dark pool participation of informed traders.
Proposition 4. In a market with an exchange and a dark pool that implements a
double auction, there exists a unique threshold volatility ¯ σ(x) > 0 with the property
that, for any σ ≤ ¯ σ(x), there exists an equilibrium (β = 0, α
d
= α

d
, α
e
= 1−α

d
), where
α

d
∈ (0, G(1)] and µ

I
solve
_
G
−1

d
) −
xS
σ
_
· (1 − ¯ r
x
) =
µ
I
µ
I
+ (1 −α
d

z
, (36)
µ
I
= ¯ µF
_
(1 −α
d

z
µ
I
+ (1 −α
d

z
σ
_
. (37)
In this equilibrium with a fixed x:
1. If c ∈ [0, xS), a liquidity buyer (resp. seller) with a delay cost of c quotes a limit
20
For tractability reasons, I have not characterized an equilibrium in which some informed traders send orders
to the limit-order dark pool. A modeling challenge with informed participation in the limit-order dark pool is to
calculate the expected loss of liquidity traders, conditional on order execution at each possible price in the interval
[−xS, xS], not only the midpoint. Boulatov and George (2010) model a nondisplayed market in which informed
traders submit demand schedules (i.e. limit orders). Their model is tractable partly because their uninformed
traders are noise traders and hence do not internalize the costs of trading against informed traders. By contrast,
endogenous venue selection of liquidity traders is a key modeling objective of this paper.
25
price of c (resp. −c) in the dark pool. If c ∈ [xS, G
−1


d
)σ], then a liquidity buyer
(resp. seller) with a delay cost of c quotes a limit price of xS (resp. −xS) in the
dark pool. Liquidity traders with delay costs higher than G
−1


d
)σ trade on the
exchange.
2. The dark pool execution price is given by (79) in the appendix.
3. The dark pool participation rate α
d
of liquidity traders, the mass µ
I
of informed
traders, and the scaled exchange spread S/σ are all strictly increasing in the value
σ of information.
Moreover, for x ∈ (0, 1), the volatility threshold ¯ σ(x) is strictly decreasing in x.
Proof. See Appendix C.
The equilibrium of Proposition 4 with a limit-order dark pool is qualitatively sim-
ilar to the equilibrium characterized in Case 1 of Proposition 1. The equilibrium is
determined by the marginal liquidity trader who is indifferent between the two venues,
shown in (36), and the marginal for-profit trader who is indifferent about whether
to acquire the information, shown in (37). If multiple equilibria exist, I select the
equilibrium with the lowest α

d
among those with the property that, as α
d
varies in a
neighborhood of α

d
, the left-hand side of (36) crosses the right-hand side from below.
The expressions of ¯ σ(x) and p

in equilibrium are provided in Appendix C.
Naturally, in equilibrium liquidity traders who have higher delay costs submit more
aggressive orders (i.e. buy orders with higher limit prices and sell orders with lower
limit prices). Moreover, because a liquidity buyer’s order is infinitesimal and has zero
impact on the execution price p

, she wishes to use a “truth-telling” strategy, that is,
to submit a buy order whose limit price is equal to her delay cost.
21
If her decay cost
c < xS, the trade-at rule is not binding, so she submit a dark pool buy order with
the limit price c. If c ≥ xS, the trade-at rule becomes binding at the price xS, so the
liquidity buyer selects the highest limit price allowed, xS. In equilibrium, a strictly
positive mass of liquidity buyers set the limit price xS and are rationed with a positive
probability. When the delay cost c is sufficiently high, the liquidity buyer trades on the
exchange in order to avoid the risk of being rationed at the price xS. The intuition for
a liquidity seller is symmetric.
Proposition 4 further reveals that the trade-at rule has a material effect for the
participation of informed traders in the dark pool. Because ¯ σ(x) is decreasing in x,
21
This strategy is reminiscent of the truth-telling strategy of MacAfee (1992), who considers a double auction
with finitely many buyers and sellers. The double auction here has the institutional restriction that transaction
prices are bounded by the trade-at rule.
26
the stricter is the trade-at rule, the less attractive is the dark pool to informed traders.
The intuition is as follows. If an informed buyer were to deviate to the dark pool, she
would select the most aggressive permissible limit price, xS, in order to maximize her
execution probability. Although she would be rationed at the price xS, she would only
compete with those liquidity traders who have a delay cost of xS or higher. The lower
is x, the less scope there is for the informed trader to “step ahead of the queue” and
gain execution priority. In particular, a midpoint dark pool with x = 0 has the greatest
effectiveness in discouraging informed traders to participate.
The left-hand plot of Figure 7 shows the dark pool orders in the equilibrium of
Proposition 4. In this example, x = 0.8, so the dark pool provides a price improvement
equal to 20% of the exchange spread S. In this example, about 95% of liquidity traders
in the dark pool set the most aggressive limit price, ±xS. The dark pool transaction
price in this case is about 0.007. The right-hand plot of Figure 7 shows that the
volatility threshold ¯ σ(x) is strictly decreasing in x. With midpoint crossing (x = 0),
informed traders avoid the dark pool if the value σ of information is lower than about
0.35. Under the current trade-through rule (x = 1), this volatility threshold is reduced
to about 0.22.
The effect of the trade-at rule on informed participation in dark pools complements
prior fairness-motivated arguments, which suggest that displayed orders should have
strictly higher priority than nondisplayed orders at the same price (Joint CFTC-SEC
Advisory Committee, 2011). Proposition 4 predicts that implementing a trade-at rule
is likely to reduce informed participation in dark pools. It also predicts that dark pools
operating as limit order books are more likely to attract informed traders and impatient
liquidity traders than dark pools crossing at the midpoint.
My model of a limit-order dark pool is related to and complements that of Buti,
Rindi, Wen, and Werner (2011), who study the effect of tick size for market quality.
In their model, a limit order book with a subpenny tick size is similar to a dark pool
studied in this section. Since their model allows traders to post limit orders on the
displayed market, it generates predictions on quote depths, which are not offered in
my model. On the other hand, my model focuses on asymmetric information and price
discovery, which are absent in their model. A desirable (and nontrivial) extension of
my price-discovery model is to fully allow limit orders in both the exchange and the
dark pool, and this extension is left for future research. The extensive literature on
displayed limit order books is surveyed by Parlour and Seppi (2008).
The model of this section also differs from existing studies of nondisplayed markets
that operate alone. For example, Boulatov and George (2010) model how informed
27
Figure 7: A dark pool as a nondisplayed limit order book. The left-hand plot shows the aggregate
limit orders in the dark pool, where y
+
(p) and y

(p) denotes the demand schedule and supply
schedule, respectively. The right-hand plot shows the range of σ for which the equilibrium of
Proposition 4 exists, that is, informed traders avoid the dark pool. Model parameters are those of
Figure 3. The left-hand plot also uses x = 0.8, σ = ¯ σ(0.8) = 0.236, and realizations Z
+
= 31 and
Z

= 30.
−0.01 −0.005 0 0.005 0.01
14
14.5
15
15.5
16
Limit price
M
a
s
s


y
+
(p)
y

(p)
0 0.2 0.4 0.6 0.8 1
0
0.1
0.2
0.3
0.4
0.5
σ


x
No informed participation
traders provide liquidity through limit orders when their demand schedules (i.e. limit
orders) are hidden. This nondisplayed market, they conclude, encourages informed
traders to trade more aggressively on their information, hence improving price discov-
ery, relative to a displayed market. Hendershott and Jones (2005) empirically study
price discovery for exchange-traded funds (ETFs) when Island ECN stopped display-
ing its limit orders. Since Island ECN was the dominant market for affected ETFs, it
differed from today’s equity dark pools, which operate alongside exchanges.
5 Dynamic Trading
This short section generalizes the basic intuition of Section 3 to a dynamic market.
Under natural conditions, all equilibria have the property that, after controlling for
delay costs, an informed trader prefers the exchange to the dark pool, relative to a
liquidity trader.
Time is discrete, t ∈ {1, 2, 3, . . . }. As before, an asset pays an uncertain dividend v
that is +σ or −σ with equal probabilities. The dividend is announced at the beginning
of period T ≥ 2, where T is deterministic, and paid at the end of period T. The trading
game ends immediately after the dividend payment.
In each period before the dividend payment, a new set of informed traders and
liquidity traders arrive. To simplify the analysis, I drop endogenous information ac-
28
quisition in this section. (Equivalently, it costs zero to acquire information.) The mass
of informed traders arriving in period t, µ
I
(t) > 0, is deterministic. Informed traders
observe the dividend v and trade in the corresponding direction. The mass of liquidity
buy orders and the mass of liquidity sell orders arriving in period t are Z
+
(t) > 0 and
Z

(t) > 0, respectively, with commonly known probability distributions. The public
does not observe v or the realizations of Z
+
(t) or Z

(t).
As before, a lit exchange and a dark pool operate in parallel. Both venues are open
in all periods. At the beginning of period t, the exchange market maker posts a bid
price B
t
and an ask price A
t
. Any order sent to the exchange is immediately executed
at the bid or the ask. After execution of exchange orders in each period, the market
maker announces the exchange buy volume and the exchange sell volume. The public
information F
t
at the beginning of period t consists of all exchange announcements
prior to, but not including, period t. Thus, the conditional distribution of asset value
v at the beginning of period t is represented by the likelihood ratio
R
t
=
P
t
(v = +σ)
P
t
(v = −σ)
, (38)
where P
t
denotes the conditional probability based on F
t
. By construction, R
0
= 1.
The public’s conditional expectation of the asset value at the beginning of period t is
therefore
V (R
t
) = σ(P
t
(v = +σ) −P
t
(v = −σ)) =
R
t
−1
R
t
+ 1
σ. (39)
The dark pool executes orders in each period, simultaneously with the execution
of exchange orders. The dark pool implements a double auction with a trade-at rule,
as in Section 4. Midpoint crossing, which offers a price improvement of the exchange
spread, is a special case of this double auction. Liquidity traders differ from each other
in their delay costs, as in Section 3. If a liquidity trader of cost type γ does not trade
in period t, then she incurs a delay cost of c(γ; R
t
) in period t, where c(γ; R
t
) is strictly
increasing in γ for all R
t
. A trader only incurs delay costs after she arrives.
To control for traders’ characteristics other than information, I make the additional
assumption that informed traders also incur positive delay costs, before they execute
their orders. A type-γ informed trader incurs the delay cost c(γ; R
t
) in period t if she
fails to execute her order in that period. In practice, this cost may come from the
opportunity cost of capital. Thus, a type-γ informed buyer (resp. seller) and a type-γ
liquidity buyer (resp. seller) differ only in their information about v.
I now fix a cost type γ ≥ 0 and compare the venue choice of a type-γ informed
29
buyer with that of a type-γ liquidity buyer. For any (R
t
, t), I let
W
e
(R
t
, t) = σ −A
t
(40)
X
e
(R
t
, t) = V (R
t
) −A
t
(41)
be the payoffs of a type-γ informed buyer and a type-γ liquidity buyer, respectively, for
trading immediately on the exchange. These payoffs do not depend on the cost type
γ because exchange execution incurs no delays. I let W
d
(R
t
, t; γ) and X
d
(R
t
, t; γ) be
the corresponding continuation values of entering an order in the dark pool. Finally,
I let W(R
t
, t; γ) and X(R
t
, t; γ) be the continuation values of the informed buyer and
liquidity buyer, respectively, at the beginning of period t, before they make trading
decisions. For t = T, W(R
T
, T; γ) = X(R
T
, T; γ) = 0. For t < T, the Bellman
Principal implies that
W(R
t
, t; γ) = max [W
e
(R
t
, t), W
d
(R
t
, t), E
t
(W(R
t+1
, t + 1; γ))] , (42)
X(R
t
, t; γ) = max [X
e
(R
t
, t), X
d
(R
t
, t), E
t
(X(R
t+1
, t + 1; γ))] , (43)
where the three terms in the max( · ) operator represents a trader’s three choices: send-
ing her order to the exchange, sending her order to the dark pool, and delaying trade.
The following proposition characterizes equilibrium conditions under which, con-
trolling for delay costs, the liquidity-versus-informed payoff difference X
d
(R
t
, t; γ) −
W
d
(R
t
, t; γ) in the dark pool is at least as high as the corresponding payoff difference
X
e
(R
t
, t) −W
e
(R
t
, t) on the exchange. It is in this “difference-in-difference” sense that
the dark pool is more attractive to liquidity traders, and that the exchange is more
attractive to informed traders.
Proposition 5. In any equilibrium, if W
d
(R
t
, t; γ) ≥ E
t
[W(R
t+1
, t + 1; γ)], then
X
d
(R
t
, t; γ) −W
d
(R
t
, t; γ) ≥ X
e
(R
t
, t) −W
e
(R
t
, t). (44)
Proof. See Appendix C.
Proposition 5 reveals that all equilibria must satisfy the restriction (44), provided
that a informed buyer weakly prefers using the dark pool to delaying trade. The
intuition for this result is as follows. Because the exchange guarantees to execute all buy
orders at the same price A
t
, the exchange payoff difference, X
e
(R
t
, t)−W
e
(R
t
, t), reflects
only the value of private information. The dark pool payoff difference X
d
(R
t
, t; γ) −
W
d
(R
t
, t; γ), by contrast, reflects both the value of information and the execution risk.
30
Compared with a liquidity buyer, an informed buyer in the dark pool is less likely
to fill her order and, conditional on a trade, more likely to pay a higher price. This
execution risk is costly for the informed buyer in equilibrium as long as she prefers
dark pool trading to delaying, as captured by W
d
(R
t
, t; γ) ≥ E
t
[W
d
(R
t+1
, t + 1; γ)].
I isolate this dark pool execution risk from the value of information by taking the
“difference-in-difference” of payoffs in (44).
Appendix B explicitly solves a dynamic equilibrium in a setting where traders arrive
in Poisson times.
6 Implications and Discussions
This section discusses some implications of my results, both in light of recent empirical
evidence and in relation to the current policy debate on the impacts of dark pools on
price discovery and liquidity. The discussion follows two organizing questions. First,
what are the relations between dark pool market share and observable market charac-
teristics? Second, what are the impacts of dark pool trading on price discovery and
liquidity? For each question, I discuss empirical implications of the model and put
them in the context of related empirical evidence.
6.1 Determinants of dark pool market share
Prediction 1. All else equal, dark pool market share is lower if the execution probability
of dark pool orders is lower.
Prediction 2. All else equal, if the level of adverse selection (or volatility) is low, then
dark pool market share is increasing in adverse selection (or volatility). If the level of
adverse selection (or volatility) is high, then dark pool market share can be decreasing
in adverse selection (or volatility).
Prediction 3. All else equal, informed participation in dark pools is higher if volatility
is higher. Informed participation is higher in dark pools that allow more discretion in
execution prices, compared with dark pools that execute orders at the exchange midpoint.
Prediction 4. All else equal, dark pool market share is lower for trading strategies
relying on shorter-term information. The use of dark pools is also lower for trading
strategies that trade multiple stocks simultaneously, compared with strategies that trade
individual stocks one at a time.
31
Prediction 1 follows from the results of Section 3 and Section 4. A lower execution
probability, captured by a lower ¯ r or ¯ r
x
, discourages both types of traders from partici-
pating in the dark pool.
22
Using daily data collected by SIFMA from eleven anonymous
dark pools in 2009, Buti, Rindi, and Werner (2011a) find that dark pool market share
is negatively related to the order imbalance as a percentage of total volume and to the
absolute depth imbalance on lit venues. Prediction 1 is consistent with their findings.
23
Prediction 2 suggests that dark pool market share can be increasing or decreasing
in the level of adverse selection (or volatility), depending on whether σ, the value of
private information, is below or above the threshold ¯ σ. Using transaction data in two
block-crossing dark pools (Liquidnet and Posit), Ready (2012) finds that institutions
are less likely to route orders to dark pools when the level of adverse selection is higher.
In different samples, Buti, Rindi, and Werner (2011a) and Ye (2010) find that dark
pool market shares are lower when volatilities and spreads are higher. To the extent
that at least some informed traders participate in dark pools in practice,
24
and that
volatilities and spreads are positively correlated with adverse selection, these findings
are broadly consistent with (the latter half of) Prediction 2.
Prediction 3 suggests that dark pool orders are more informative on average when
information asymmetry is severe. This prediction is consistent with recent evidence
documented by Nimalendran and Ray (2012) in an anonymous dark pool. They infer
the trading direction of each dark pool transaction by comparing the execution price
with the prevailing market midpoint. A trading strategy that follows the directions of
dark pool orders is profitable when spreads are wide but not profitable when spreads are
narrow. To the extent that exchange spreads are proxy measures for adverse selection,
Prediction 3 is consistent with their results. Prediction 3 also suggests that orders in
limit-order dark pools are more informative than those in midpoint dark pools. To my
knowledge, the latter half of Prediction 3 is not yet tested in the data.
Prediction 4 provides strategy-level implications on dark pool activity. Strategies
relying on shorter-term information have higher execution risks in dark pools because
relevant information can become stale sooner. Related to this prediction, Ready (2012)
finds that the usage of block-crossing dark pools is lower for institutions with higher
turnover, which is consistent with the notion that short-term strategies are best imple-
mented in venues that guarantee execution. Because dark pools cannot guarantee the
22
This relation can be analytically proved for σ < ¯ σ in Proposition 1 and for σ < ¯ σ(x) in Proposition 4.
23
Related, Ye (2010) constructs a proxy for execution probability in eight dark pools from their SEC Rule 605 re-
ports, and studies the relationship between non-execution probability and market characteristics (e.g. price impacts
and effective spreads). He does not examine how non-execution probability relates to the market share of dark pools.
For more details of Rule 605 of Reg NMS, see http://www.finra.org/Industry/Regulation/Guidance/SECRule605/.
24
Recall from the model of Section 3 that if σ > ¯ σ, then some informed traders participate in the dark pool.
32
simultaneous execution of trades in multiple stocks, we also expect dark pools to be
less attractive for strategies tracking stock indices or “arbitraging” perceived mispric-
ing among similar securities. For these strategies, partial execution in dark pools can
be particularly costly.
6.2 Effects of dark pools on price discovery and liquidity
Prediction 5. All else equal, a higher dark pool market share is associated with higher
order informativeness, wider spreads, and higher price impacts of trades on the ex-
change.
Prediction 6. All else equal, a higher dark pool market share increases the correlation
of volumes across different stocks in lit exchanges. This cross-stock volume correlation
is lower in dark pools than in lit exchanges.
Prediction 7. All else equal, dark pool execution implies a positive adverse-selection
cost, in that shares bought in dark pools tend to have low short-term returns and that
shares sold in dark pools tend to have high short-term returns. This cost, however, is
lower than the exchange spread at the time of execution.
Proposition 3 provides sufficient conditions under which Prediction 5 holds. Pre-
diction 5 is consistent with empirical evidence from Degryse, de Jong, and van Kervel
(2011), Nimalendran and Ray (2012), Jiang, McInish, and Upson (2011), and Weaver
(2011), but not Buti, Rindi, and Werner (2011a) or O’Hara and Ye (2011). In Dutch eq-
uity markets, Degryse, de Jong, and van Kervel (2011) find that higher market shares
of dark trading—including dark pools and over-the-counter markets—are associated
with higher price impacts, higher quoted spreads, higher realized spreads, and smaller
depths on lit markets. Similarly, in U.S. equity markets, Jiang, McInish, and Upson
(2011) find that off-exchange (dark) order flows are less informative than exchange (lit)
order flows, after adjusting for trading volumes in dark and lit markets. Their results
also indicate that exchange order flows become more informative as off-exchange order
flows increase. Weaver (2011) finds that higher levels of off-exchange trading in the
U.S. are associated with wider spreads, higher price impacts, and higher volatilities.
Using transaction data in an anonymous dark pool, Nimalendran and Ray (2012) doc-
ument that following dark-pool transactions, bid-ask spreads tend to widen and price
impacts tend to increase, especially if the relative bid-ask spreads are already high. By
contrast, Buti, Rindi, and Werner (2011a) find that higher dark pool trading activity
tends to be associated with lower spreads and lower return volatilities, which suggest a
33
better market quality. O’Hara and Ye (2011) also conclude that higher fragmentation
of trading is associated with faster execution, lower transaction costs, and more efficient
prices. Given the wide variety of data samples used in these studies and the difficulty
in completely correcting for endogeneity, we should interpret these conflicting results
with caution.
Prediction 6 can be viewed as the mirror image of Prediction 4. Since dark pools are
less attractive to strategies that execute multiple stocks simultaneously, those strategies
should have higher concentration in lit venues than in dark pools. Consequently, the
volume correlation across stocks should be higher in lit venues than in dark pools. To
my knowledge, this prediction is not yet tested in the data.
Finally, Prediction 7 on the adverse selection in dark pools is consistent with Sofianos
and Xiang (2011), who find that dark pools that have higher execution probabilities
also have more severe adverse selection (that is, more “toxic”). Næs and Odegaard
(2006) provide anecdotal evidence that filled orders in a dark pool are subject to short-
term losses. Mittal (2008) and Saraiya and Mittal (2009) emphasize that short-term
adverse selection in dark pools can reduce execution quality of institutional investors.
Conrad, Johnson, and Wahal (2003), Brandes and Domowitz (2010), and Domowitz,
Finkelshteyn, and Yegerman (2009) examine execution costs in dark pools, although
they do not explicitly measure the costs of adverse selection.
7 Concluding Remarks
In recent years, dark pools have become an important part of equity market structure.
This paper provides a simple model of dark pool trading and their effects on price
discovery and liquidity. I show that under natural conditions, the addition of a dark
pool concentrates informed traders on the exchange and improves price discovery, at
the cost of reducing exchange liquidity.
Besides price discovery and liquidity, there are a few additional aspects of dark pools
that contribute to their controversy. One of these is information leakage. In practice,
a dark pool may send an “indication of interest” (IOI), which contains selected order
information such as the ticker, to potential counterparties in order to facilitate a match.
In this sense, these dark pools are not completely dark. For example, Buti, Rindi, and
Werner (2011b) considers a setting where selected traders are informed of the state
of the dark pool. The Securities and Exchange Commission (2009) proposed to treat
actionable IOIs—IOIs containing the symbol, size, side, and price of an order—as
quotes, which must be disseminated to the broad market immediately.
34
Another consideration is fair access. In the United States, dark pools are not re-
quired to provide fair access unless the dark pool concerned reaches a 5% volume
threshold. Whether investors suffer from the lack of fair access can depend on perspec-
tive. On the one hand, it seems plausible that the lack of fair access can reduce trading
opportunities and the welfare of excluded traders. On the other hand, “some dark
pools attempt to protect institutional trading interest by raising access barrier to the
sell-side or certain hedge funds,” observes SEC Deputy Director James Brigagliano.
25
For example, results from Boni, Brown, and Leach (2012) indicate that the exclusion of
short-term traders in a dark pool (Liquidnet) improves the execution quality of insti-
tutional orders. Foster, Gervais, and Ramaswamy (2007) theoretically illustrate that
setting a volume threshold in the dark pool—i.e. the dark pool executes orders only
if trading interests on both sides of the market reach that threshold—can sometimes
prevent impatient traders or informed traders from participating in the dark pool.
Finally, dark pools are opaque not only in their orders, but also in their trading
mechanisms. For example, a Greenwich Associates survey of 64 active institutional
users of dark pools reveals that, on many occasions, dark pools do not disclose suffi-
cient information regarding the types of orders that are accepted, how orders interact
with each other, how customers’ orders are routed, what anti-gaming controls are in
place, whether customer orders are exposed to proprietary trading flows, and at what
price orders are matched (Bennett, Colon, Feng, and Litwin, 2010). The International
Organization of Securities Commissions (2010) also observes that “[l]ack of information
about the operations of dark pools and dark orders may result in market participants
making uninformed decisions regarding whether or how to trade within a dark pool or
using a dark order.” Opaque operating mechanics of dark pools can make it more diffi-
cult for investors and regulators to evaluate the impact of dark pools on price discovery,
liquidity, and market quality.
25
“Keynote Speech to the National Organization of Investment Professionals,” by James A. Brigagliano, April
19, 2010.
35
Appendix
A Institutional Features of Dark Pools
This appendix discusses additional institutional features of dark pools and nondisplayed
liquidity that are not covered in Section 2.
Besides the three-way classification of dark pools discussed in Section 2, another
classification is provided by Tabb Group (2011). They categorize dark pools into
block-cross platforms, continuous-cross platforms, and liquidity-provider platforms.
The main features of these three groups are summarized in the top panel of Table 2,
and their respective market shares are plotted in Figure 8. As we can see, the market
share of block-cross dark pools has declined from nearly 20% in 2008 to just above
10% in 2011. Continuous-cross dark pools have gained market share during the same
period, from around 50% to around 70%. The market share of liquidity-providing dark
pools increased to about 40% around 2009, but then declined to about 20% in mid
2011. Tabb Group’s data, however, do not cover the entire universe of dark pools, and
the components of each category can vary over time. For this reason, these statistics
are noisy and should be interpreted with caution.
Figure 8: Market shares of three types of U.S. dark pools as fractions of total U.S. dark pool
volume, estimated by Tabb Group. The three types are summarized in the top panel of Table 2.


2009 2010 2011
0
0.2
0.4
0.6
0.8
1
Block cross
Continuous cross
Liquidity providers
Dark pools are also commonly classified by their crossing frequencies and by how
they search for matching counterparties, as illustrated in the bottom panel of Table 2.
36
T
a
b
l
e
2
:
D
a
r
k
p
o
o
l
s
c
l
a
s
s
i

c
a
t
i
o
n
b
y
t
r
a
d
i
n
g
m
e
c
h
a
n
i
s
m
s
.
T
h
e
t
o
p
p
a
n
e
l
s
h
o
w
s
t
h
e
c
l
a
s
s
i

c
a
t
i
o
n
o
f
T
a
b
b
G
r
o
u
p
.
T
h
e
b
o
t
t
o
m
p
a
n
e
l
s
h
o
w
s
a
c
l
a
s
s
i

c
a
t
i
o
n
b
y
t
h
e
c
r
o
s
s
i
n
g
f
r
e
q
u
e
n
c
i
e
s
a
n
d
t
h
e
m
e
t
h
o
d
s
o
f

n
d
i
n
g
c
o
u
n
t
e
r
p
a
r
t
i
e
s
.
C
l
a
s
s
i

c
a
t
i
o
n
b
y
T
a
b
b
G
r
o
u
p
T
y
p
e
s
E
x
a
m
p
l
e
s
N
o
t
e
s
B
l
o
c
k
c
r
o
s
s
L
i
q
u
i
d
n
e
t
,
B
I
D
S
,
I
n
s
t
i
n
e
t
C
r
o
s
s
S
i
m
i
l
a
r
t
o
t
h
e

r
s
t
g
r
o
u
p
o
f
t
h
e
t
o
p
p
a
n
e
l
C
o
n
t
i
n
u
o
u
s
c
r
o
s
s
C
r
e
d
i
t
S
u
i
s
s
e
C
r
o
s
s

n
d
e
r
,
G
o
l
d
m
a
n
S
a
c
h
s
S
i
g
m
a
X
,
B
a
r
c
l
a
y
s
L
X
,
M
o
r
g
a
n
S
t
a
n
l
e
y
M
S
P
o
o
l
,
L
e
v
e
L
,
D
e
u
t
s
c
h
e
B
a
n
k
S
u
p
e
r
X
S
i
m
i
l
a
r
t
o
t
h
e
s
e
c
o
n
d
g
r
o
u
p
o
f
t
h
e
t
o
p
p
a
n
e
l
;
L
e
v
e
L
i
s
o
w
n
e
d
b
y
a
c
o
n
s
o
r
t
i
u
m
o
f
b
r
o
k
e
r
-
d
e
a
l
e
r
s
L
i
q
u
i
d
i
t
y
p
r
o
v
i
d
e
r
G
e
t
c
o
a
n
d
K
n
i
g
h
t
S
a
m
e
a
s
t
h
e
t
h
i
r
d
g
r
o
u
p
o
f
t
h
e
t
o
p
p
a
n
e
l
C
l
a
s
s
i

c
a
t
i
o
n
b
y
t
r
a
d
i
n
g
f
r
e
q
u
e
n
c
y
a
n
d
c
o
u
n
t
e
r
p
a
r
t
y
s
e
a
r
c
h
T
y
p
e
s
E
x
a
m
p
l
e
s
T
y
p
i
c
a
l
f
e
a
t
u
r
e
s
S
c
h
e
d
u
l
e
d
I
T
G
P
O
S
I
T
M
a
t
c
h
,
I
n
s
t
i
n
e
t
U
S
C
r
o
s
s
i
n
g
C
r
o
s
s
a
t

x
e
d
c
l
o
c
k
t
i
m
e
s
,
w
i
t
h
s
o
m
e
r
a
n
d
o
m
i
z
a
-
t
i
o
n
C
o
n
t
i
n
u
o
u
s
M
a
t
c
h
i
n
g
I
T
G
P
O
S
I
T
N
o
w
,
I
n
s
t
i
n
e
t
C
B
X
,
D
i
r
e
c
t
E
d
g
e
M
i
d
P
o
i
n
t
M
a
t
c
h
A
d
v
e
r
t
i
z
e
d
P
i
p
e
l
i
n
e
,
P
O
S
I
T
A
l
e
r
t
,
L
i
q
u
i
d
n
e
t
E
l
e
c
t
r
o
n
i
c
m
e
s
s
a
g
e
s
s
e
n
t
t
o
p
o
t
e
n
t
i
a
l
m
a
t
c
h
e
d
c
o
u
n
t
e
r
p
a
r
t
i
e
s
N
e
g
o
t
i
a
t
e
d
L
i
q
u
i
d
n
e
t
I
n
t
e
r
n
a
l
C
r
e
d
i
t
S
u
i
s
s
e
C
r
o
s
s

n
d
e
r
,
G
o
l
d
m
a
n
S
a
c
h
s
S
i
g
m
a
X
,
K
n
i
g
h
t
L
i
n
k
,
G
e
t
c
o
O
w
n
e
d
b
y
b
r
o
k
e
r
-
d
e
a
l
e
r
s
,
r
u
n
a
s
n
o
n
d
i
s
p
l
a
y
e
d
l
i
m
i
t
o
r
d
e
r
b
o
o
k
s
o
r
e
l
e
c
t
r
o
n
i
c
m
a
r
k
e
t
m
a
k
e
r
s
37
Aside from mechanisms such as midpoint-matching and limit order books, advertise-
ment is sometimes used to send selected information about orders resting in the dark
pool to potential counterparties, in order to facilitate a match.
Characteristics that distinguish dark pools also include ownership structure and
order size. Today, most dark pools are owned by broker-dealers (with or without pro-
prietary order flows), whereas a small fraction is owned by consortiums of broker-dealers
or exchanges. Order sizes can also vary substantially across dark pools. According to
Rosenblatt Securities (2011), two block-size dark pools (Liquidnet and Pipeline) have
an order size of around 50,000 shares, which is larger than that of Posit (around 6,000
shares per order) and much larger than those of other broker dark pools (about 300
shares per order). This sharp contrast in order sizes can be attributed to the use of
algorithms that split “parent” orders into smaller “children” orders, as observed by the
Securities and Exchange Commission (2010).
There are at least two reasons why high-quality data are lacking on dark pool
trading in the United States. First, in the United States, dark pool trades are reported
to “trade reporting facilities,” or TRFs, which aggregate trades executed by all off-
exchange venues—including dark pools, ECNs, and broker-dealer internalization—into
a single category. Thus, it is generally not possible to assign a TRF trade to a specific
off-exchange venue that executes the trade.
26
Second, dark pools often do not have their
own identification numbers (MPID) for trade reporting. For example, a broker-dealer
may report customer-to-customer trades in its dark pool together with the broker’s own
over-the-counter trades with institutions, all under the same MPID. Similarly, trades
in an exchange-owned dark pool can be reported together with trades conducted on the
exchange’s open limit order book, all under the exchange’s MPID. Because different
trading mechanisms share the same MPID, knowing the MPID that executes a trade
is insufficient to determine whether that trade occurred in a dark pool.
27
Finally, there are two sources of nondisplayed liquidity that are usually not referred
to as dark pools. One is broker-dealer internalization, by which a broker-dealer handles
customer orders as a principal or an agent (Securities and Exchange Commission, 2010).
A crude way of distinguishing dark pools from broker-dealer internalization is that the
former are often marketplaces that allow direct customer-to-customer trades, whereas
the latter typically involves broker-dealers as intermediaries.
28
The other source of
26
The Securities and Exchange Commission (2009) has recently proposed a rule requiring that alternative trading
systems (ATS), including dark pools, provide real-time disclosure of their identities on their trade reports.
27
For example, Ye (2010) finds that only eight U.S. dark pools can be uniquely identified by MPIDs from their
Rule 605 reports to the SEC. The majority of dark pools cannot.
28
There are exceptions. For example, dark pools acting like electronic market makers (like Getco and Knight)
also provide liquidity by trading on their own accounts. Nonetheless, they are highly automated systems and rely
38
nondisplayed liquidity is the use of hidden orders on exchanges. Examples include
reserve (“iceberg”) orders and pegged orders, which are limit orders that are partially or
fully hidden from the public view.
29
For example, Nasdaq reports that more than 15%
of its order flow is nondisplayed.
30
In particular, midpoint-pegged orders on exchanges
are similar to dark pool orders waiting to be matched at the midpoint.
B Dynamic Trading with Stochastic Crossing
This appendix explicitly characterizes a family of dynamic equilibria in which informed
traders do not participate in the dark pool. Different from Section 5, information and
traders in this section arrive at Poisson times, which give rise to tractable stationary
equilibria.
Time is continuous, t ≥ 0, and the market opens at time 0. As before, an asset
pays an uncertain dividend v that is +σ or −σ with equal probabilities. The time of
the dividend payment is exponentially distributed with mean 1/λ
F
, for λ
F
> 0. Two
types of risk-neutral traders—liquidity traders and informed traders—have independent
Poisson arrivals with respective mean arrival rates of λ
L
and λ
I
. (Traders are thus
“discrete.”) Each trader can buy or sell one unit of the asset. As in Section 5, I do not
consider endogenous information acquisition here. Upon arrival, an informed trader
observes v perfectly. Liquidity traders, who are not informed regarding the dividend,
arrive with an unwanted position in the asset whose size is either +1 or −1, equally
likely and independent of all else.
As before, a lit exchange and a dark pool operate in parallel. A competitive and
risk-neutral market maker on the exchange continually posts bid and ask prices for one
unit of the asset, as in Glosten and Milgrom (1985). Any order sent to the exchange
is immediately executed at the bid or the ask, and trade information is immediately
disseminated to everyone. By competitive pricing, the bid price at any time t is the
conditional expected asset value given the arrival of a new sell order at time t and
given all public information up to, but before, time t. The ask price is set likewise.
The market maker also maintains a public “midpoint” price that is the conditional
expected asset value given all public information up to but before time t. Once an
less on human intervention than, say, dealers arranging trades over the telephone.
29
A reserve order consists of a displayed part, say 200 shares, and a hidden part, say 1,800 shares. Once the
displayed part is executed, the same amount, taken from the hidden part, becomes displayed, until the entire order
is executed or canceled. Pegged orders are often fully hidden. Typically, pegged orders and hidden portions of
reserve orders have lower execution priority than displayed orders with the same limit price.
30
See http://www.nasdaqtrader.com/Trader.aspx?id=DarkLiquidity
39
exchange order is executed, the market maker immediately updates her bid, ask, and
midpoint prices.
The dark pool accepts orders continually, and an order sent to the dark pool is
observable only by the order submitter. The dark pool executes orders at the midpoint
price and at the event times of a Poisson process with intensity λ
C
that is independent of
all else. Allocation in the dark pool is pro-rata on the heavier side, as in Section 3. For
analytical tractability, I assume that unmatched orders in the dark pool are immediately
sent to the exchange market maker, who then executes these orders at the conditional
expected asset value given all past public information and given the quantity and
direction of unmatched orders from the dark pool.
As in Section 5, the conditional likelihood ratio of v at time t is
R
t
=
P
t
(v = +σ)
P
t
(v = −σ)
, (45)
where P
t
denotes the market maker’s conditional probability. By construction, R
0
= 1.
The conditional expected asset value is, as in Section 5,
V (R
t
) = σ(P
t
(v = +σ) −P(v = −σ)) =
R
t
−1
R
t
+ 1
σ. (46)
To calculate the bid and ask prices, I let λ
t
be the time-t arrival intensity (conditional
mean arrival rate) of traders of any type to the exchange, and let µ
t
be the time-t
conditional probability that an arriving exchange trader is informed. Then,
q
t
= µ
t
+ (1 −µ
t
)0.5 = 0.5 + 0.5µ
t
(47)
is the probability that an exchange trader arriving at t is “correct,” that is, buying if
v = +σ and selling if v = −σ. The likelihood ratio
z
t
=
q
t
1 −q
t
(48)
then represents the informativeness of a time-t exchange order.
31
For example, if a buy
order hits the market maker’s bid at time t, then Bayes’ Rule implies that
R
t
=
P
t
(v = +σ | Q = 1)
P
t
(v = −σ | Q = 1)
=
P
t
(Q = 1 | v = +σ)
P
t
(Q = 1 | v = −σ)
·
P
t−
(v = +σ)
P
t−
(v = −σ)
= R
t−
z
t
, (49)
31
In the equilibria characterized in this section, the information content of a buy order is equal to that of a sell
order, so there is no need to specify them separately.
40
where P
t−
denotes the market maker’s probability conditional on all exchange transac-
tions up to but before time t, and where R
t−
≡ lim
s↑t
R
s
. Similarly, if an exchange sell
order arrives at time t, then
R
t
=
P
t
(v = +σ | Q = −1)
P
t
(v = −σ | Q = −1)
=
R
t−
z
t
. (50)
To break even, the market maker quotes a time-t bid price of V (R
t
z
−1
t
) and a time-t
ask price of V (R
t
z
t
). Because V ( · ) is nonlinear, V (R
t
) is generally not identical to the
bid-ask midpoint, (V (R
t
z
t
) +V (R
t
z
−1
t
))/2. Nonetheless, for simplicity I refer to V (R
t
)
as the “midpoint” price.
Liquidity traders must hold collateral equal to the expected loss on their unwanted
risky positions. With probability κ
j
and independently of all else, an arriving liquidity
trader incurs a cost of γ
j
per unit of time for every unit of collateral support in her
risky position, where (κ
j
)
J
j=1
and (γ
j
)
J
j=1
are commonly-known constants and satisfy
0 ≤ γ
1
< γ
2
< · · · < γ
J−1
< γ
J
, (51)
J

j=1
κ
j
= 1. (52)
Before executing her order, a liquidity buyer of type j incurs a flow cost of
c
j
t
= γ
j
E
t
[max(0, v −V (R
t
))] = γ
j
R
t
R
t
+ 1
·
_
1 −
R
t
−1
R
t
+ 1
_
σ = γ
j
2R
t
(R
t
+ 1)
2
σ. (53)
A liquidity seller of type j has the same flow cost c
j
t
because
γ
j
E
t
[max(0, V (R
t
) −v)] = γ
j
1
R
t
+ 1
·
_
R
t
−1
R
t
+ 1
−(−1)
_
σ = γ
j
2R
t
(R
t
+ 1)
2
σ. (54)
By independent splitting of Poisson processes, the arrival intensities of type-j liquidity
buyers and type-j liquidity sellers are both 0.5κ
j
λ
L
.
Without loss of generality, we focus on the strategies of informed buyers and liquidity
buyers, whose payoffs are denoted W(R
t
) and X(R
t
), respectively. For simplicity, I use
E
i
t
[ · ] as a shorthand for E
t
[ · | v = σ], where the superscript “i” stands for “informed.”
Because I look for stationary equilibria, the payoffs W(R
t
) and X(R
t
) depend on the
public information R
t
but not on time t.
41
Proposition 6. For fixed integer M ∈ {0, 1, 2, . . . , J}, define
z
e
=
λ
I
+ 0.5

J
i=M
κ
i
λ
L
0.5

J
i=M
κ
i
λ
L
. (55)
Under the conditions
λ
C
<

J
i=M
κ
j
λ
L

I
λ
F
, (56)
γ
j
< (λ
C

F
)
z
e
−1
z
e
, 1 ≤ j < M, (57)
γ
j
> (λ
C

F
)(z
e
−1) +
_
λ
I
+
J

i=M
κ
i
λ
L
_
(z
e
−1)
3
z
e
(

z
e
+ 1)
2
, M ≤ j ≤ J, (58)
there exists an equilibrium in which:
1. Informed traders trade on the exchange immediately upon arrival.
2. Type-j liquidity traders, M ≤ j ≤ J, trade immediately on the exchange upon
arrival.
3. Type-j liquidity traders, 1 ≤ j < M, enter orders in the dark pool. If the dark
pool has not crossed by the time that the dividend is paid, they cancel their dark
pool orders and trade immediately on the exchange.
4. At time t, the market maker quotes a bid of V (R
t
z
−1
e
) and an ask of V (R
t
z
e
).
Moreover, immediately after a dark pool crossing, the market maker executes all
outstanding orders at a price of V (R
t
). Immediately after the dividend v is paid,
the market maker executes all outstanding orders at the cum-dividend price of v.
Proof. See Appendix C.
A key step in the equilibrium solution of Proposition 6 is that informed traders
expect the exchange price to move against them over time, but liquidity traders expect
the exchange spread to narrow over time. Thus, informed traders are relatively impa-
tient, whereas liquidity traders are relatively patient. These different expectations of
future prices, as formally stated in the following lemma, underlie the partial separation
between informed traders and liquidity traders in the equilibria of Proposition 6.
Lemma 1. Let Q be the direction of the next exchange order that arrives before the
dividend payment, that is, Q = 1 denotes a buy order and Q = −1 denotes a sell order.
Under the strategies stated in Proposition 6:
42
• The asset value is a martingale for liquidity traders and the public, in that
V (R
t
) = E
t
[V (R
t
z
Q
e
)]. (59)
• The exchange ask price is a submartingale for informed buyers, in that
V (R
t
z
e
) < E
i
t
[V (R
t
z
Q
e
z
e
)]. (60)
• The exchange ask price is a supermartingale for liquidity buyers, in that
E
t
[V (R
t
z
Q
e
z
e
)] = V (R
t
z
e
) −
2R
2
t
(z
e
−1)
3
(R
t
+ 1)
2
(R
t
z
e
+ 1)(R
t
z
2
e
+ 1)
σ. (61)
Proof. See Appendix C.
In Proposition 6, z
e
reflects the degree of information asymmetry on the exchange
because it is the ratio of the mean arrival rate of traders in the “correct” direction versus
the mean arrival rate of traders in the “wrong” direction. Proposition 6 says that an
informed trader trades immediately on the exchange if the crossing frequency λ
C
of the
dark pool is sufficiently low relative to the risk that her private information becomes
stale. A liquidity trader sends her order to the dark pool if and only if her delay cost
γ is sufficiently low compared to the potential price improvement obtained by trading
at the market midpoint. Moreover, because the exchange order informativeness z
e
is
increasing in M, the more liquidity traders trade in the dark pool, the more informative
are exchange orders. This property is a dynamic analogue of the two-period equilibrium
of Section 3.
We now briefly discuss the comparative statics of the equilibria, based on the tight-
ness of the incentive constraints (56)-(58). First, a higher crossing frequency λ
C
tightens
(56), suggesting that informed traders are more likely to participate in the dark pool
if the crossing frequency is higher. On the other hand, a higher λ
C
relaxes (57) but
tightens (58), making the dark pool more attractive to liquidity traders. As long as
λ
C
is sufficiently low, informed traders avoid in the dark pool. Second, a higher arrival
rate λ
F
of information relaxes (56), suggesting that informed traders are less likely to
trade in the dark pool if they face a higher risk of losing their information advantage.
By contrast, a higher λ
F
makes the dark pool more attractive to liquidity traders by
shortening their expected waiting time, as in (57)-(58). Third, a higher delay cost γ
makes the dark pool less attractive to liquidity traders, without affecting the incentives
of informed traders.
43
C Proofs
C.1 Proof of Proposition 1
I define ˆ µ
I
: [0, ∞) →[0, ¯ µ] by
ˆ µ
I
(s) = ¯ µF
_
(1 −G(1))µ
z
ˆ µ
I
(s) + (1 −G(1))µ
z
s
_
. (62)
Given the value σ of information, ˆ µ
I
(σ) is the unique “knife-edge” mass of informed
traders with the property that all informed traders and a fraction 1 −G(1) of liquidity
traders send orders to the exchange.
To prove the proposition, I show that a Case 1 equilibrium exists if
¯ r ≤ 1 −
ˆ µ
I
(σ)
ˆ µ
I
(σ) + (1 −G(1))µ
z
, (63)
and that a Case 2 equilibrium exists if and only if
¯ r > 1 −
ˆ µ
I
(σ)
ˆ µ
I
(σ) + (1 −G(1))µ
z
. (64)
Then I show that the condition (63) is equivalent to σ ≤ ¯ σ for some ¯ σ, and that the
condition (64) is equivalent to σ > ¯ σ.
Clearly, β < 1; otherwise, the exchange spread would be zero and informed traders
would deviate to trade on the exchange. Thus, in equilibrium either β = 0 or 0 < β < 1.
We first look for an equilibrium in which β = 0. By (15), α
0
= 0 and α
e
=
1−α
d
. The indifference condition of the marginal liquidity trader is given by (16). For
notational simplicity, we write the left-hand side of (16) as −
˜
X
d

d
) and the right-hand
side as −
˜
X
e

d
). For each α
d
, µ
I
is uniquely determined by (17). We have

˜
X
d
(0) = 0 < −
˜
X
e
(0),

˜
X
d
(G(1)) = 1 − ¯ r ≥
ˆ µ
I
(σ)
ˆ µ
I
(σ) + (1 −G(1))µ
z
= −
˜
X
e
(G(1)),
where the second inequality follows from (63), (17), and (62). So there exists a solution
α

d
∈ (0, G(1)] that satisfies (16).
Now we look for an equilibrium in which β > 0, that is, informed traders are
indifferent between the exchange and the dark pool. What remains to be shown is
that the incentive-compatibility conditions (18)-(20) have a solution. For simplicity,
we write the left-hand side of (18) as
˜
W
d
(β) and the right-hand side of (18) as
˜
W
e
(β).
44
For each β ≥ 0, µ
I
is unique determined by (20) and is increasing in β. Under condition
(64) and for each α
d
> 0,
˜
W
d
(0) = ¯ r > 1 −
ˆ µ
I
(σ)
ˆ µ
I
(σ) + (1 −G(1))µ
z
=
˜
W
e
(0),
˜
W
d
(1) = r

< 1 =
˜
W
e
(1),
where the first inequality follows from (63), (17), and (62). So there exists a solution
β

∈ (0, 1) to (18), as a function of α
d
. Because µ
I
increases in β, we see that
˜
W

d
(β) < 0
and
˜
W

e
(β) > 0, holding α
d
fixed. Thus, the solution β

to (18) is unique for each α
d
.
Moreover, (18) implies that in equilibrium r

is bounded away from 0. So there
exists some r
0
> 0 such that r

> r
0
. So for sufficiently small α
d
> 0,
G(1) −G
_
r
+
−r

r
+
+r

_
> G(1) −G
_
1 −r
0
1 +r
0
_
> α
d
.
So there exists a solution α

d
∈ (0, G(1)] to (19). The equilibria characterized by (18)-
(20) thus exist. To show that (64) is necessary for the existence of equilibria in which
β > 0, suppose for contradiction that (64) does not hold. Then, for all α
d
and β > 0,
˜
W
e
(β) >
˜
W
e
(0) ≥
˜
W
d
(0) >
˜
W
d
(β), which implies that all informed traders wish to
deviate to the exchange, contradicting β > 0.
Finally, by (62), increasing the value σ of information raises the knife-edge mass ˆ µ(σ)
of informed traders, which in turn tightens the condition (63) under which informed
traders avoid the dark pool. Thus, there exists some unique volatility threshold ¯ σ at
which (63) holds with an equality. That is, the equilibrium in Case 1 exists if σ ≤ ¯ σ,
and the equilibrium in Case 2 exists if σ > ¯ σ.
C.2 Proof of Proposition 2
Because β, α
d
, α
e
, µ, S, r
+
and r

are implicitly defined by differentiable functions in
each case of Proposition 1, they are continuous and differentiable in σ in each of the
two intervals [0, ¯ σ] and (¯ σ, ∞). At the volatility threshold σ = ¯ σ, differentiability refers
to right-differentiability in Case 1 of Proposition 1, and left-differentiability in Case 2.
45
Have a dark pool and σ ≤ ¯ σ
For σ ≤ ¯ σ, β = 0. Total differentiation of (16)-(17) with respect to σ yields
_
dG
−1

d
)

d
(1 − ¯ r) −
∂(S/σ)
∂α
d
_
. ¸¸ .
>0

d


∂(S/σ)
∂µ
I
. ¸¸ .
>0

I

= 0, (65)
_
1 − ¯ µF

(σ −S)
∂ (σ −S)
∂µ
I
_
. ¸¸ .
>0

I

= ¯ µF

(σ −S)
∂(σ −S)
∂α
d
. ¸¸ .
<0

d

+ ¯ µF

(σ −S)
_
1 −
S
σ
_
. ¸¸ .
>0
,
(66)
where the first term of (65) is positive because of equilibrium selection. If dα
d
/dσ ≤ 0
at, say, some σ
0
, then (66) implies that dµ
I
/dσ > 0 at σ
0
. But then (65) cannot hold.
Thus, dα
d
/dσ > 0, dµ
I
/dσ > 0, and d(S/σ)/dσ > 0, by (16).
Have a dark pool and σ > ¯ σ
Now suppose that σ > ¯ σ. I denote by r
+

and r

the derivatives of r
+
and r

with
respect to βµ
I

d
. We have r
+

> 0 and r

< 0. Total differentiation of (18)-(20)
with respect to σ yields
_
r

1
α
d

∂(1 −S/σ)
∂(βµ
I
)
_
. ¸¸ .
<0
d(βµ
I
)

=
∂(1 −S/σ)
∂µ
I
. ¸¸ .
<0

I

+r

βµ
I
α
2
d
. ¸¸ .
<0

d

, (67)
_
1 −G

_
r
+
−r

r
+
+r

_
2(r
+

r

−r

r
+
)
(r
+
+r

)
2
βµ
I
α
2
d
_
. ¸¸ .
>0

d

= −G

_
r
+
−r

r
+
+r

_
2(r
+

r

−r

r
+
)
(r
+
+r

)
2
1
α
d
. ¸¸ .
>0
d(βµ
I
)

,
(68)
_
1 − ¯ µF

(σ −S)σ
∂(1 −S/σ)
∂µ
I
_
. ¸¸ .
>0

I

= ¯ µF

(σ −S)σ
∂(1 −S/σ)
∂(βµ
I
)
. ¸¸ .
>0
d(βµ
I
)

+ ¯ µF

(σ −S)
_
1 −
S
σ
_
. ¸¸ .
>0
,
(69)
where the first term of (68) is positive because of equilibrium selection.
We can show that dα
d
/dσ cannot switch signs in [¯ σ, ∞). To see why, suppose
otherwise, and dα
d
/dσ switches signs at some σ
0
. By continuity, at σ
0
, dα
d
/dσ = 0.
But (68) and (67) imply that d(βµ
I
)/dσ = 0 = dµ
I
/dσ at σ
0
as well, which contradicts
(69). Thus, dα
d
/dσ cannot switch signs in [¯ σ, ∞); nor can it be zero.
46
At σ = ¯ σ, β = 0 and dβ/dσ ≥ 0. Then, by (68),
d(βµ
I
)

¸
¸
¸
σ=¯ σ
= µ
I


¸
¸
¸
σ=¯ σ
≥ 0 =⇒

d

¸
¸
¸
σ=¯ σ
≤ 0.
Because dα
d
/dσ cannot be zero, it must be strictly negative for all σ ∈ [¯ σ, ∞). By
(68)-(69), for all σ ∈ [¯ σ, ∞), βµ
I
and µ
I
are both strictly increasing in σ. Then, (18)
implies that
d(S/σ)

= −
dr


= −r

d

_
βµ
I
α
d
_
> 0.
The spread itself, S = σ · (S/σ), obviously increases in σ as well. Finally,
dr
+

= r
+
d

_
βµ
I
α
d
_
> 0,
dr


= r

d

_
βµ
I
α
d
_
< 0.
No dark pool
The comparative statics for Corollary 1 are similar to that for the first case of Propo-
sition 1 and are omitted.
C.3 Proof of Proposition 3
Have a dark pool and σ ≤ ¯ σ
For σ ≤ ¯ σ, adding a dark pool is equivalent to increasing ¯ r. Total differentiation of
(16)-(17) with respect to ¯ r yields
_
(1 − ¯ r)
∂G
−1

d
)
∂α
d

∂(S/σ)
∂α
d
_
. ¸¸ .
>0

d
d¯ r
= G
−1

d
) +
∂(S/σ)
∂µ
I
. ¸¸ .
>0

I
d¯ r
, (70)
_
1 − ¯ µF

(σ −S)

∂µ
I
(σ −S)
_
. ¸¸ .
>0

I
d¯ r
= ¯ µF

(σ −S)
∂(σ −S)
∂α
d
. ¸¸ .
<0

d
d¯ r
, (71)
where the first term on the left-hand side of (70) is positive because of the equilibrium
selection. If dα
d
/d¯ r ≤ 0 at any σ
0
, then (71) implies that dµ
I
/d¯ r ≥ 0 at σ
0
. But that
contradicts (70). Thus, dα
d
/d¯ r > 0 and dµ
I
/d¯ r < 0. Adding a dark pool, which is
equivalent to an increase in ¯ r, raises α
d
and reduces α
e
= 1−α
d
. The total participation
rate of liquidity traders in either the dark pool or the exchange is α
d
+ α
e
= 1, higher
than a market without a dark pool. Moreover, by (17), a lower µ
I
implies a wider
spread S on the exchange.
47
Have a dark pool and σ > ¯ σ
Now suppose that σ > ¯ σ. In a market with a dark pool, α
e
= 1 − G(1), a constant.
Substituting it into (21) and we have
µ
I
µ
I
+ (1 −G(1))µ
z
< 1.
So the equilibrium α
e
without a dark pool resides in the interval (1 − G(1), 1). That
is, adding a dark pool reduces α
e
.
Moreover, adding a dark pool increases the exchange spread if and only if α
e
in the
equilibrium of Corollary 1 is larger than (1 −G(1))/(1 −β), where β > 0 is determined
in Proposition 1. By the equilibrium selection rule and by (18),
α
e
>
1 −G(1)
1 −β
⇐⇒G
−1
_
1 −
1 −G(1)
1 −β
_
>
µ
I
µ
I

z
(1 −G(1))/(1 −β)
= 1 −r

,
(72)
where the µ
I
is given by
µ
I
= ¯ µF
_
(1 −G(1))µ
z
(1 −β)µ
I
+ (1 −G(1))µ
z
_
.
We rearrange (72) and obtain
β <
G(1) −G(1 −r

)
1 −G(1 −r

)
.
On the other hand, because the left-hand side of (18) is decreasing in β and the right-
hand side is increasing in β, the above condition is equivalent to (23).
As F(c) → 1 for all c > 0, (20) implies that µ
I
→ ¯ µ, a constant. Holding µ
I
= ¯ µ
fixed, we now show that if G

(1 − r

) ≤ 0, then (23) holds for all r

∈ [0, ¯ r]. At
r

= ¯ r, we have σ = ¯ σ and (23) holds by the definition of ¯ σ. At r

= 0, (23) also
holds trivially. Take the first and second derivatives of the right-hand side of (23) with
respect to r

and we obtain
d[rhs(23)]
dr

=
¯ µµ
z
G

(1 −r

)
[¯ µ + (1 −G(1 −r

))µ
z
]
2
> 0,
d
2
[rhs(23)]
d(r

)
2
= ¯ µµ
z
G

(1 −r

)[¯ µ + (1 −G(1 −r

))µ
z
] −2µ
z
[G

(1 −r

)]
2
[¯ µ + (1 −G(1 −r

))µ
z
]
3
< 0.
Thus, the right-hand side of (23) is concave and (23) holds for all r

∈ [0, ¯ r].
48
C.4 Proof of Proposition 4
I prove this proposition in three steps. First, I calculate the execution price in the dark
pool and the optimal limit prices chosen by liquidity traders. Second, I derive incentive-
compatibility conditions under which informed traders choose not to participate in the
dark pool. Finally, I prove the comparative statics and conditions under which the
equilibrium exists.
Step 1: Price p

of execution and optimal prices of limit orders
I let y
+
: [−xS, xS] → [0, ∞) be the aggregate downward-sloping demand schedule of
liquidity buyers in the dark pool, and let y

: [−xS, xS] → [0, ∞) be the aggregate
upward-sloping supply schedule of liquidity sellers. For each p, y
+
(p) is the total mass
of limit buy orders that have a limit price of p or higher, and y

(p) is the total mass
of limit sell orders that have a limit price of p or lower. Because the dark pool crosses
orders by price priority, its execution price p

is
p

=
_
¸
¸
¸
_
¸
¸
¸
_
xS, if y
+
(p) > y

(p) for all p ∈ [−xS, xS].
−xS, if y
+
(p) < y

(p) for all p ∈ [−xS, xS].
{p : y
+
(p) = y

(p)}, otherwise.
(73)
I proceed under the conjecture that the set {p : y
+
(p) = y

(p)} contains at most one
element, in which case p

of (73) is uniquely well-defined. I later verify this conjecture.
Once p

is determined, buy orders with limit prices above or equal to p

are matched,
at the price of p

, with sell orders whose prices are at most p

. If there is a positive
mass of buy or sell orders at the price p

, then traders setting the limit price p

are
rationed pro-rata, as before.
I now derive the optimal limit prices of liquidity traders in the dark pool, under
the conjecture that the probability distribution of p

has no atom in (−xS, xS). This
no-atom conjecture, verified later, implies that a liquidity trader quoting a price of
p ∈ (−xS, xS) has her order filled with certainty (i.e. is not rationed) if p

= p. Thus,
a liquidity buyer who has a delay cost of c ∈ [0, xS) and quotes a price of p in the dark
pool has the expected payoff (negative cost)
X
d
(p; c) = −E
_
I
{p≥p

}
p

+I
{p<p

}
c
¸
= −c −
_
p
−xS
(p

−c) dH(p

), (74)
49
where I( · ) is the indicator function and H(p

) is the cumulative distribution function
of p

. Because there is no adverse selection in the dark pool, the execution cost for this
liquidity buyer is either the payment p

or the delay cost c. Conjecturing that H(p

)
is differentiable with H

(p

) > 0 for p

∈ (−xS, xS), properties that are also verified
later, we obtain
dX
d
(p; c)
dp
= −(p −c)H

(p). (75)
Because (75) shows that the sensitivity of expected payoff to the limit price p is positive
for p < c and negative for p > c, the optimal limit price for the liquidity buyer is her
delay cost c. Symmetrically, the optimal limit price for a liquidity seller with a delay
cost of c ∈ [0, xS) is −c. This “truth-telling” strategy is also ex-post optimal, in
that no one wishes to deviate even after observing the execution price. The first-order
condition (75) also implies that xS is the highest limit price in the dark pool, and that
−xS is the lowest limit price.
32
Let y(p) be the downward-sloping demand schedule in the dark pool if Z
+
= 1.
Because a limit price p ∈ [0, xS) is submitted by the liquidity buyer with the delay
cost p,
y(p) = α
d
−G
_
max(0, p)
σ
_
, −xS < p < xS. (76)
By symmetry, the liquidity buyers’ demand schedule and the liquidity sellers’ supply
schedule in the dark pool are, respectively,
y
+
(p) = Z
+
y(p), (77)
y

(p) = Z

y(−p). (78)
Because the equation y
+
(p) = y

(p) has at most one root, we have verified our earlier
conjecture that the dark pool execution price p

is uniquely well-defined.
Given y(p), the execution price p

in the dark pool is
p

=
_
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
_
+xS, if [α
d
−G
_
xS
σ
_
]Z
+
≥ α
d
Z

,
+σG
−1
_
α
d
_
1 −
Z

Z
+
__
, if
_
α
d
−G
_
xS
σ

Z
+
< α
d
Z

≤ α
d
Z
+
,
−σG
−1
_
α
d
_
1 −
Z
+
Z

__
, if
_
α
d
−G
_
xS
σ

Z

< α
d
Z
+
≤ α
d
Z

,
−xS, if [α
d
−G
_
xS
σ
_
]Z

≥ α
d
Z
+
.
(79)
32
If the maximum limit price were lower, say p
0
< xS, then a liquidity buyer with a delay cost of p
0
+ for some
small > 0 would deviate to the dark pool and quote p
0
+. This deviating buyer has an execution probability of
1 and pays at most p
0
+ < xS ≤ S, which is better than execution on the exchange. The argument for the lowest
limit price is symmetric.
50
Because the total trading interest Z
+
of liquidity buyer and the total trading interest
Z

of liquidity sellers are identically distributed, the dark pool execution price p

has a
mean of zero. By the differentiability of G and of the distribution function of Z

/Z
+
,
H(p

) is continuous, differentiable, and strictly increasing on (−xS, xS), as conjectured
earlier.
Step 2: Incentive conditions for participation
What remains to be shown are the incentive-compatibility conditions of liquidity traders
who set the limit price xS or −xS in the dark pool, as well as the incentive-compatibility
condition of informed traders, who avoid the dark pool. A liquidity buyer quoting the
limit price xS in the dark pool has an execution probability of
¯ r
x
= E
_
min
_
1,
α
d
Z


d
−G(xS/σ))Z
+
__
, (80)
and an expected payoff, given the delay cost c, of
X
d
(xS; c) = −(1 − ¯ r
x
)(c −xS). (81)
This expected payoff calculation follows from the fact that E(p

) = 0 and the fact that
failing to cross in the dark pool incurs a delay cost of c but saves the payment xS.
Because informed traders avoid the dark pool with probability 1 in the conjectured
equilibrium, an informed buyer who deviates to the dark pool also has the crossing
probability ¯ r
x
. Moreover, in order to get the highest priority, this deviating informed
trader sets the highest limit price xS. Her expected profit in the dark pool is thus
W
d
= σ −(1 − ¯ r
x
)(σ −xS). (82)
As before, for any delay cost c ≤ σ,
W
d
−X
d
(xS; c) = σ¯ r
x
+c(1 − ¯ r
x
) ≤ σ = W
e
−X
e
. (83)
That is, an informed buyer behaves in the same way as does a liquidity buyer who has a
delay cost of σ. If informed traders do not participate in the dark pool, an equilibrium
is determined by a marginal liquidity trader who is indifferent between the dark pool
and the exchange. Given α
d
, this liquidity trader has a delay cost of G
−1

d
)σ. So
we must have X
d
(xS; G
−1

d
)) = −S, or (36). Thus, (36) and (37) characterize an
equilibrium. We now look for conditions under which, in equilibrium, α

d
≤ G(1). In
51
this equilibrium, β = 0.
Step 3: Comparative statics and conditions for the existence of equilibria
I now calculate the comparative statics, assuming the existence of an equilibrium, and
then show conditions under which the stated equilibrium exists. Total differentiation
of (36) and (37) with respect to σ yields
_
∂[lhs(36)]
∂α
d

∂[rhs(36)]
∂α
d
_
. ¸¸ .
>0

d

+
_
∂[lhs(36)]
∂µ
I

∂[rhs(36)]
∂µ
I
_
. ¸¸ .
<0

I

= 0, (84)
_
1 −
∂[rhs(37)]
∂µ
I
_
. ¸¸ .
>0

I

=
∂[rhs(37)]
∂α
d
. ¸¸ .
<0

d

+ ¯ µF

(σ −S)
_
1 −
S
σ
_
. ¸¸ .
>0
. (85)
As before, if dα
d
/dσ ≤ 0 at some σ
0
, then (84) implies that dµ
I
/dσ ≤ 0 at σ
0
as well.
But this contradicts (85). Thus, the comparative statics with respect to σ follow. And
given the equilibrium, the dark pool execution price p

and the optimal limit prices
follow from calculations done in Step 1 of the proof.
Now I characterize the condition for the existence of an equilibrium and the thresh-
old volatility ¯ σ(x). For x ∈ [0, 1], I define
¯
K(x) implicitly by
(1 −x
¯
K(x))
_
1 −E
_
min
_
1,
G(1)Z

[G(1) −G(x
¯
K(x))]Z
+
___
=
¯
K(x). (86)
This
¯
K(x) is uniquely well-defined because the left-hand side of (86) is decreasing in
¯
K(x) and the right-hand side is strictly increasing in
¯
K(x). Moreover, total differenti-
ation of (86) with respect to x yields
_
∂[lhs(86)]

¯
K(x)
−1
_
. ¸¸ .
<0
¯
K

(x) +
∂[lhs(86)]
∂x
. ¸¸ .
<0
= 0.
So we have
¯
K

(x) < 0.
On the other hand, given
¯
K(x), I define µ

I
(x) by
µ

I
(x)
µ

I
(x) + (1 −G(1))µ
z
=
¯
K(x),
52
and define ¯ σ(x) by
µ

I
(x) = ¯ µF
_
(1 −G(1))µ
z
µ

I
(x) + (1 −G(1))µ
z
¯ σ(x)
_
.
Because µ

I
(x) is strictly increasing in
¯
K(x) and because ¯ σ(x) is strictly increasing in
µ

I
(x), ¯ σ(x) is strictly increasing in
¯
K(x). Because
¯
K

(x) < 0, ¯ σ

(x) < 0.
What remains to be shown is that, for σ ≤ ¯ σ(x), an equilibrium characterized by
Proposition 4 exists. Clearly, once α
d
is determined, µ
I
is uniquely determined by (37).
For sufficiently small α
d
, the left-hand side of (36) is negative, whereas the right-hand
side is strictly positive. For α
d
= G(1), (37) implies that
µ
I
= ¯ µF
_
(1 −G(1))µ
z
µ
I
+ (1 −G(1))µ
z
σ
_
,
which is no larger than µ

I
(x). Thus,
K ≡
µ
I
µ
I
+ (1 −G(1))µ
z
=
S
σ
¸
¸
¸
α
d
=G(1)

¯
K(x),
and, by the definition of
¯
K(x),
(1 −xK)
_
1 −E
_
min
_
1,
G(1)Z

[G(1) −G(xK)]Z
+
___
> K.
That is, at α
d
= G(1), the left-hand side of (36) is weakly higher than the right-hand
side. Therefore, there exists a solution α

d
∈ (0, G(1)) to (36), and an equilibrium
exists.
C.5 Proof of Proposition 5
Suppose that W
d
(R
t
, t; γ) ≥ E
t
[W(R
t+1
, t + 1; γ)] in an equilibrium. I denote by
ˆ
X(R
t
, t; γ) the “auxiliary payoff” of a type-γ liquidity buyer who “imitates” the strat-
egy of a type-γ informed buyer. That is, the imitating buyer behaves as if v = +σ.
Clearly, such imitation is suboptimal for the liquidity buyer, so X(R
t
, t; γ) ≥
ˆ
X(R
t
, t; γ)
for all R
t
, t, and γ. For notional simplicity, in the calculations below I suppress the
cost type γ and likelihood ratio R
t
as function arguments.
Suppose that the informed buyer enters an order in the dark pool at time t. The
imitating liquidity buyer does the same, by construction. Let
ˆ
X
+
d
be the dark pool
payoff of the imitating buyer conditional on v = +σ, and let
ˆ
X

d
be the dark pool
53
payoff of the imitating buyer conditional on v = −σ. I define
ˆ
X
+
e
and
ˆ
X

e
similarly.
Then, we have
ˆ
X
d
(t) −W
d
(t) =
R
t
R
t
+ 1
ˆ
X
+
d
(t) +
1
R
t
+ 1
ˆ
X

d
(t) −W
d
(t)
=
1
R
t
+ 1
_
ˆ
X

d
(t) −
ˆ
X
+
d
(t)
_
,
where the last equality follows from the fact that, conditional on the true dividend, the
expected payoff W
d
of the informed buyer is the same as the payoff
ˆ
X
+
d
of the imitating
liquidity buyer. If we can show that
ˆ
X
+
d
(t) −
ˆ
X

d
(t) ≤ 2σ, then
X
d
(t) −W
d
(t) ≥
ˆ
X
d
(t) −W
d
(t) ≥ −

R
t
+ 1
= X
e
(R
t
, t) −W
e
(R
t
, t).
Now I prove that
ˆ
X
+
d
(t) −
ˆ
X

d
(t) ≤ 2σ. I denote by C the event that the imitating
buyer’s order is crossed in the dark pool, and let
k
+
t
≡ P
t
[C | v = +σ],
k

t
≡ P
t
[C | v = −σ],
be the crossing probabilities of the imitating buyer in the dark pool in period t, condi-
tional on v = +σ and v = −σ, respectively. Then, we have
ˆ
X
+
d
(t) = k
+
t
[+σ −E
t
(p

| C, v = +σ)] + (1 −k
+
t
)E
t
(
ˆ
X
+
(t + 1)),
ˆ
X

d
(t) = k

t
[−σ −E
t
(p

| C, v = −σ)] + (1 −k

t
)E
t
(
ˆ
X

(t + 1)),
where p

is the execution price in the dark pool in period t, and
ˆ
X
+
and
ˆ
X

are the
imitating buyer’s payoffs conditional on v = +σ and v = −σ, respectively. Informed
buyers have either a zero or positive mass in the dark pool in period t, so
k
+
t
≤ k

t
,
E
t
(p

| C, v = +σ) ≥ E
t
(p

| C, v = −σ).
Because W
d
(t) ≥ E
t
[W(t + 1)],
ˆ
X
+
d
(t) ≥ E
t
(
ˆ
X
+
(t + 1)). Thus,
ˆ
X
+
d
(t) −E
t
(
ˆ
X
+
(t + 1)) = k
+
t
[σ −E
t
(p

| C, v = +σ) −E
t
(
ˆ
X
+
(t + 1))]
≤ k

t
[σ −E
t
(p

| C, v = −σ) −E
t
(
ˆ
X
+
(t + 1))],
54
which implies that
ˆ
X
+
d
(t) −
ˆ
X

d
(t) ≤ 2σk

t
+ (1 −k

t
)E
t
[
ˆ
X
+
(t + 1) −
ˆ
X

(t + 1)]. (87)
I now prove that
ˆ
X
+
d
(t) −
ˆ
X

d
(t) ≤ 2σ and that
ˆ
X
+
(t) −
ˆ
X

(t) ≤ 2σ by induction.
For all t < T, X
+
e
(t) − X

e
(t) = 2σ. Because v is revealed in period T,
ˆ
X
+
(T) =
ˆ
X

(T) = 0. By (87),
ˆ
X
+
d
(T −1) −
ˆ
X

d
(T −1) ≤ 2σ. Because the venue choice of the
imitating liquidity buyer does not depend on realizations of v,
X
+
(T −1) −X

(T −1)
= max
_
X
+
e
(T −1) −X

e
(T −1), X
+
d
(T −1) −X

d
(T −1), E
T−1
(X
+
(T) −X

(T))
¸
≤ 2σ.
For the induction step, suppose that
ˆ
X
+
(t + 1) −
ˆ
X

(t + 1) ≤ 2σ. Then, (87) implies
that X
+
d
(t) −X

d
(t) ≤ 2σ. Thus,
X
+
(t) −X

(t) = max
_
X
+
e
(t) −X

e
(t), X
+
d
(t) −X

d
(t), E
t
(X
+
(t + 1) −X

(t + 1))
¸
≤ 2σ,
which completes the proof.
C.6 Proof of Lemma 1
Given the public information R
t
, the probability that v = +σ is R
t
/(R
t
+ 1). Let q
be implicitly defined by z
e
= q/(1 − q). That is, q is the probability that an arriving
exchange order is in the same direction as informed orders. Under a liquidity trader’s
belief, the probability that the next exchange order is a buy order is
R
t
R
t
+ 1
q +
1
R
t
+ 1
(1 −q) = (1 −q)
R
t
z
e
+ 1
R
t
+ 1
.
Similarly, the probability that the next exchange order is a sell order is
R
t
R
t
+ 1
(1 −q) +
1
R
t
+ 1
q = q
R
t
z
−1
e
+ 1
R
t
+ 1
.
55
We can verify that identity
(1 −q)
R
t
z
e
+ 1
R
t
+ 1
V (R
t
z
e
) +q
R
t
z
−1
e
+ 1
R
t
+ 1
V (R
t
z
−1
e
)
=(1 −q)
R
t
z
e
+ 1
R
t
+ 1
_
1 −
2
R
t
z
e
+ 1
_
σ +q
R
t
z
−1
e
+ 1
R
t
+ 1
_
1 −
2
R
t
z
−1
e
+ 1
_
σ
=
_
1 −
2
R
t
+ 1
_
σ, (88)
which implies E
t
[V (R
t
z
Q
e
)] = V (R
t
), that is, the expected midpoint price is a martingale
for liquidity traders.
Note that the identity (88) holds for all R
t
, including the informed trader’s likelihood
ratio ∞. Using (88) again, we have that
V (R
t
z
e
) =
_
1 −
2
R
t
z
e
+ 1
_
σ
= (1 −q)
R
t
z
e
z
e
+ 1
R
t
z
e
+ 1
_
1 −
2
R
t
z
e
z
e
+ 1
_
σ +q
R
t
z
e
z
−1
e
+ 1
R
t
z
e
+ 1
_
1 −
2
R
t
z
e
z
−1
e
+ 1
_
σ
< (1 −q)z
e
_
1 −
2
R
t
z
e
z
e
+ 1
_
σ +qz
−1
e
_
1 −
2
R
t
z
e
z
−1
e
+ 1
_
σ
= E
i
t
[V (R
t
z
Q
e
z
e
)].
That is, the exchange ask price is a submartingale for informed buyers.
Finally, direct calculation gives
E
t
[V (R
t
z
Q
e
z
e
)] = V (R
t
z
e
) −
2R
2
t
(z
e
−1)
3
(R
t
+ 1)
2
(R
t
z
e
+ 1)(R
t
z
2
e
+ 1)
σ,
E
t
[V (R
t
z
Q
e
z
−1
e
)] = V (R
t
z
−1
e
) +
2R
2
t
(1 −z
−1
e
)
3
(R
t
+ 1)
2
(R
t
z
−1
e
+ 1)(R
t
z
−2
e
+ 1)
σ.
That is, for liquidity traders, the exchange ask price is a supermartingale and the
exchange bid price is a submartingale.
C.7 Proof of Proposition 6
I prove Proposition 6 by direct verification. The quoting strategy of the market maker
simply follows from risk neutrality and zero profit.
Under the proposed equilibrium strategy, the total arrival intensity of exchange
orders is λ
t
= λ
I
+

J
i=M
κ
i
λ
L
. The Hamilton-Jacobi-Bellman (HJB) equation of an
56
informed buyer is
W(R
t
) = max
_
σ −V (R
t
z
e
),
λ
t
E
i
t
[W(R
t
z
Q
e
)] +λ
C
(σ −V (R
t
))
λ
t

C

F
_
,
where the profit of immediate trading on exchange is σ − V (R
t
z
e
) and the expected
profit of trading in the dark pool is the weighted sum of:
• E
i
t
[W(R
t
z
Q
e
)], the expected profit if the next event is the arrival of an exchange
order.
• σ −V (R
t
), the profit if the next event is a dark pool cross.
• 0, the profit if the next event is the dividend payment.
To verify that W(R
t
) = σ − V (R
t
z
e
), it is sufficient to verify that, for all t and all
realizations of random variable R
t
,
σ −V (R
t
z
e
) >
λ
t
E
i
t
[σ −V (R
t
z
Q
e
z
e
)] +λ
C
(σ −V (R
t
))
λ
t

C

F
. (89)
By Lemma 1, the expected profit for informed buyers to trade on the exchange is a
supermartingale, that is, σ−V (R
t
z
e
) > E
i
t
[σ−V (R
t
z
Q
e
z
e
)]. Thus, a sufficient condition
for (89) is
λ
C

F
> sup
R∈(0,∞)
_
λ
C
σ −V (R)
σ −V (Rz
e
)
_
= z
e
λ
C
,
which simplifies to (56).
Now we turn to a type-j liquidity buyer, whose HJB equation is
X(R
t
) = max
_
−(V (R
t
z
e
) −V (R
t
)),
λ
t
E
t
[X(R
t
z
Q
e
)] −c
j
t
λ
t

C

F
_
.
Her cost of liquidation C(R
t
) satisfies the HJB equation
C(R
t
) = −X(R
t
) = min
_
V (R
t
z
e
) −V (R
t
),
λ
t
E
t
[C(R
t
z
Q
e
)] +c
j
t
λ
t

C

F
_
.
There are two cases, depending on j.
If 1 ≤ j < M, to verify that C(R
t
) < V (R
t
z
e
) −V (R
t
), it suffices to verify
V (R
t
z
e
) −V (R
t
) >
λ
t
E
t
[V (R
t
z
Q
e
z
e
) −V (R
t
z
Q
e
)] +c
j
t
λ
t

C

F
, (90)
where C(R
t
z
Q
e
) is replaced by the higher cost of V (R
t
z
Q
e
z
e
) − V (R
t
z
Q
e
), as implied by
57
the conjectured equilibrium. Because the ask spread is a supermartingale for liquidity
buyers (Lemma 1), a sufficient condition for (90) is
λ
C

F
> sup
R∈(0,∞)
_
c
j
t
V (Rz
e
) −V (R)
_
= sup
R∈(0,∞)
_
γ
j
· 2R/(R + 1)
2
· σ
V (Rz
e
) −V (R)
_
= γ
j
z
e
z
e
−1
,
which simplifies to (57).
If M ≤ j ≤ J, to verify that C(R
t
) = V (R
t
z
e
) −V (R
t
), it suffices to verify
V (R
t
z
e
) −V (R
t
) <
λ
t
E
t
[V (R
t
z
Q
e
z
e
) −V (R
t
z
Q
e
)] +c
j
t
λ
t

C

F
,
that is, by Lemma 1,

C

F
)[V (R
t
z
e
) −V (R
t
)] < −λ
t
2R
2
t
(z
e
−1)
3
(R
t
+ 1)
2
(R
t
z
e
+ 1)(R
t
z
2
e
+ 1)
σ +γ
j
2R
t
(R
t
+ 1)
2
σ.
(91)
A sufficient condition for (91) is
γ
j
> sup
R∈(0,∞)
_

C

F
)
(R + 1)(z
e
−1)
Rz
e
+ 1
_
+ sup
R∈(0,∞)
_
λ
t
R(z
e
−1)
3
(Rz
e
+ 1)(Rz
2
e
+ 1)
_
= (λ
C

F
)(z
e
−1) +λ
t
(z
e
−1)
3
z
e
(

z
e
+ 1)
2
.
The argument for sellers is symmetric and yields the same parameter conditions.
58
D List of Model Variables
This appendix summarizes key variables used in Section 3 and Section 4.
Variable Description
Variables Introduced in Section 3
v, σ Asset value v is either +σ or −σ, for σ > 0
¯ µ, µ
I
Total masses of for-profit traders and informed traders
F Cumulative distribution function (c.d.f.) of information-acquisition cost
Y Signed informed trading interests: Y = sign(v) · µ
I
Z
+
, Z

, φ Liquidity buy quantity Z
+
and liquidity sell quantity Z

have p.d.f. φ
µ
z
, σ
2
z
Total mean and variance of liquidity trading interests Z
+
+Z

c, γ, G Delay cost of a liquidity trader is c = σγ per unit of asset, and γ has c.d.f. G
α
e
, α
d
, α
0
Fractions of liquidity traders who trade on the exchange, trade in the dark
pool, and defer trading, respectively
β Fraction of informed traders who trade in the dark pool
S Exchange (effective) spread; bid is −S and ask is S
¯ r Dark pool crossing probability if no informed traders go to the dark pool
r

, r
+
Dark pool crossing probabilities conditional on informed traders being on the
same and opposite side, respectively
¯ σ Maximum volatility for which informed traders avoid the dark pool
ˆ µ
I
(σ) Knife-edge mass of informed traders, defined by (62)
W
e
, W
d
Expected profits of an informed buyer on the exchange and in the dark pool
X
0
(c), X
e
, X
d
(c) Per-unit payoff of a liquidity buyer with a delay cost of c who defers trading,
trades on the exchange, and trades in the dark pool, respectively
R
1
Period-1 log likelihood ratio of {v = +σ} versus {v = −σ}
P
1
Period-1 closing price on the exchange
I(β, α
e
) Signal-to-noise ratio of period-1 exchange order flow
V
b
, V
s
Period-1 realized buy volume and sell volume on the exchange, respectively
V
d
, V
e
, V Expected volumes in the dark pool, on the exchange, and both, respectively
Variables Introduced in Section 4
x Strictness of trade-at rule; maximum (minimum) dark pool price is xS (−xS)
y Aggregate demand schedule in the dark pool if Z
+
= 1
y
+
, y

Aggregate dark pool demand and supply schedules, respectively
p

, H Dark pool transaction price p

has a c.d.f. of H
X
d
(p; c) Dark pool payoff of a liquidity buyer with the limit price p and per-unit delay
cost c
¯ r
x
Dark pool crossing probability of a liquidity buyer with the limit price xS
¯ σ(x) Maximum volatility for which informed traders avoid the dark pool
59
References
Admati, A. R., and Pfleiderer, P. 1988. A Theory of Intraday Patterns: Volume and
Price Variability. Review of Financial Studies 1: 3–40.
Bennett, J., Colon, J., Feng, J., and Litwin, J. 2010. Ten Questions Every Institution
Should Ask Their Dark Pool Providers. Journal of Trading 5: 10–13.
Bond, P., Edmans, A., and Goldstein, I. 2012. The Real Effects of Financial Markets.
Forthcoming, Annual Reviews of Financial Economics.
Boni, L., Brown, D. C., and Leach, J. C. 2012. Dark Pool Exclusivity Matters. Working
paper.
Boulatov, A., and George, T. J. 2010. Securities Trading when Liquidity Providers are
Informed. Working paper University of Houston.
Brandes, Y., and Domowitz, I. 2010. Alternative Trading Systems in Europe: Trading
Performance by European Venues Post-MiFID. Journal of Trading 5: 17–30.
Brunnermeier, M. K., and Pedersen, L. H. 2005. Predatory Trading. Journal of Finance
60: 1825–1863.
Buti, S., Rindi, B., Wen, Y., and Werner, I. M. 2011. Tick Size Regulation, Intermarket
Competition and Sub-Penny Trading. Working paper.
Buti, S., Rindi, B., and Werner, I. M. 2011a. Diving into Dark Pools. Working paper
Fisher College of Business, Ohio State University.
Buti, S., Rindi, B., and Werner, I. M. 2011b. Dynamic Dark Pool Trading Strategies
in Limit Order Markets. Working paper.
Butler, G. 2007. Liquidity Aggregation: What Institutional Investors Need to Know.
Journal of Trading 2: 108–113.
Carlin, B. I., Lobo, M. S., and Viswanathan, S. 2007. Episodic Liquidity Crisis: Coop-
erative and Predatory Trading. Journal of Finance 62: 2235–2274.
Carrie, C. 2008. Illuminating the New Dark Influence on Trading and U.S. Market
Structure. Journal of Trading 3: 40–55.
CFA Institute 2009. Market Microstructure: The Impact of Fragmentation under the
Markets in Financial Instruments Directive. Technical report CFA Institute.
60
Chordia, T., Roll, R., and Subrahmanyam, A. 2011. Recent Trends in Trading Activity
and Market Quality. Journal of Financial Economics 101: 243–263.
Chowdhry, B., and Nanda, V. 1991. Multimarket Trading and Market Liquidity. Review
of Financial Studies 4: 483–511.
Conrad, J., Johnson, K. M., and Wahal, S. 2003. Institutional trading and alternative
trading systems. Journal of Financial Economics 70: 99–134.
CSA/IIROC 2009. Dark Pools, Dark Orders, and Other Development in Market Struc-
ture in Canada. Consulation paper 23-404 Canadian Securities Administration and
Investment Industry Regulatory Organization of Canada.
Degryse, H., de Jong, F., and van Kervel, V. 2011. The Impact of Dark and Visible
Fragmentation on Market Quality. Working paper.
Degryse, H., Van Achter, M., and Wuyts, G. 2009. Dynamic Order Submission Strate-
gies with Competition between a Dealer Market and a Crossing Network. Journal of
Financial Economics 91: 319–338.
Domowitz, I., Finkelshteyn, I., and Yegerman, H. 2009. Cul de Sacs and Highways, An
Optical Tour of Dark Pool Trading Performance. Journal of Trading 4: 16–22.
Easley, D., Keifer, N. M., and O’Hara, M. 1996. Cream-Skimming or Profit-Sharing?
The Curious Role of Purchased Order Flow. Journal of Finance 51: 811–833.
European Commission 2010. Review of the Markets in Financial Instruments Directive
(MiFID). Public consultation European Commission.
Foster, D., Gervais, S., and Ramaswamy, K. 2007. The Benefits of Volume-Conditional
Order-Crossing. Working paper.
Glosten, L. R., and Milgrom, P. R. 1985. Bid, Ask and Transaction Prices in a Specialist
Market with Heterogenously Informed Traders. Journal of Financial Economics 14:
71–100.
Hendershott, T., and Jones, C. M. 2005. Island Goes Dark: Transparency, Fragmenta-
tion, and Regulation. Review of Financial Studies 18: 743–793.
Hendershott, T., and Mendelson, H. 2000. Crossing Networks and Dealer Marekts:
Competition and Performance. Journal of Finance 55: 2071–2115.
61
International Organization of Securities Commissions 2010. Issues Raised by Dark Liq-
uidity. Consultation report Technical Committee of the International Organization
of Securities Commissions.
International Organization of Securities Commissions 2011. Principles for Dark Liquid-
ity. Final report Technical Committee of the International Organization of Securities
Commissions.
Jiang, C., McInish, T., and Upson, J. 2011. Why Fragmented Markets Have Better Mar-
ket Quality: The Flight of Liquidity Order Flows to Off Exchange Venues. Working
paper.
Johnson, B. 2010. Algorithmic Trading & DMA: An introduction to direct access trading
strategies: . 4Myeloma Press, London.
Joint CFTC-SEC Advisory Committee 2011. Recommendations regarding Regulatory
Responses to the Market Event of May 6, 2010. Summary report.
Kyle, A. S. 1985. Continuous Auctions and Insider Trading. Econometrica 53: 1315–
1335.
MacAfee, R. P. 1992. A Dominant Strategy Double Auction. Journal of Economic
Theory 56: 434–450.
Mittal, H. 2008. Are You Playing in a Toxic Dark Pool? A Guide to Preventing
Information Leakage. The Journal of Trading 3: 20–33.
Næs, R., and Odegaard, B. A. 2006. Equity trading by institutional investors: To cross
or not to cross?. Journal of Financial Markets 9: 79–99.
Nimalendran, M., and Ray, S. 2012. Informational Linkages Between Dark and Lit
Trading Venues. Working paper University of Florida.
O’Hara, M., and Ye, M. 2011. Is Market Fragmentation Harming Market Quality?.
Journal of Financial Economics 100: 459–474.
Pagano, M. 1989. Trading Volume and Asset Liquidity. Quarterly Journal of Economics
104: 255–274.
Parlour, C. A., and Seppi, D. J. 2008. Limit Order Markets: A Survey. Chapter 3,
Handbook of Financial Intermediation and Banking.
62
Ready, M. J. 2012. Determinants of Volume in Dark Pools. Working paper University
of Wisconsin-Madison.
Rosenblatt Securities 2011. Let there be light. Market structure analysis.
Saraiya, N., and Mittal, H. 2009. Understanding and Avoiding Adverse Selection in
Dark Pools. Working paper Investment Technology Group.
Securities and Exchange Commission 2005. Regulation NMS. Federal Register 70:
37496–37644.
Securities and Exchange Commission 2009. Regulation of Non-Public Trading Interests;
Proposed Rule. Federal Register 74: 61207–61238.
Securities and Exchange Commission 2010. Concept Release on Equity Market Struc-
ture; Proposed Rule. Federal Register 75: 3593–3614.
Sofianos, G., and Xiang, J. 2011. Dark Pool Races, Part Two. Working paper. Goldman
Sachs.
Sun, Y. 2006. The exact law of large numbers via Fubini extension and characterization
of insurable risks. Journal of Economic Theory 126: 31–69.
Tabb Group 2011. Liquidity Matrix. Technical report.
Weaver, D. 2011. Off-Exchange Reporting and Market Quality in a Fragmented Market
Structure. Working paper.
Ye, M. 2010. Non-execution and Market Share of Crossing Networks. Working paper
University of Illinois.
Ye, M. 2011. A Glimpse into the Dark: Price Formation, Transaction Costs and Market
Share of the Crossing Network. Working paper University of Illinois.
63

1

Introduction

Dark pools are equity trading systems that do not publicly display orders. Some dark pools passively match buyers and sellers at exchange prices, such as the midpoint of the exchange bid and offer. Other dark pools essentially operate as nondisplayed limit order books that execute orders by price and time priority. In this paper, I investigate the impact of dark pools on price discovery. Contrary to misgivings expressed by some regulators and market participants, I find that under natural conditions, adding a dark pool improves price discovery on the exchange. According to the Securities and Exchange Commission (SEC; 2010), as of September 2009, 32 dark pools in the United States accounted for 7.9% of total equity trading volume. As of mid-2011, industry estimates from the Tabb Group, a consultancy, and Rosenblatt Securities, a broker, attribute about 12% of U.S. equity trading volume to dark pools. The market shares of dark pools in Europe, Canada, and Asia are smaller but quickly growing (International Organization of Securities Commissions, 2010). Dark pools have raised regulatory concerns in that they may harm price discovery. The European Commission (2010), for example, remarks that “[a]n increased use of dark pools . . . raise[s] regulatory concerns as it may ultimately affect the quality of the price discovery mechanism on the ‘lit’ markets.” The International Organization of Securities Commissions (2011) similarly worries that “the development of dark pools and use of dark orders could inhibit price discovery if orders that otherwise might have been publicly displayed become dark.” According to a recent survey conducted by the CFA Institute (2009), 71% of respondents believe that the operations of dark pools are “somewhat” or “very” problematic for price discovery. The Securities and Exchange Commission (2010), too, considers “the effect of undisplayed liquidity on public price discovery” an important regulatory question. Speaking of nondisplayed liquidity, SEC Commissioner Elisse Walter commented that “[t]here could be some truth to the criticism that every share that is crossed in the dark is a share that doesn’t assist the market in determining an accurate price.”1 My inquiry into dark pools builds on a simple model of strategic venue selection by informed and liquidity traders. Informed traders hope to profit from proprietary information regarding the value of the traded asset, whereas liquidity traders wish to meet their idiosyncratic liquidity needs. Both types of traders optimally choose between an exchange and a dark pool. The exchange displays a bid and an ask and executes all submitted orders at the bid or the ask. The dark pool free-rides on exchange prices
1

“Speech by SEC Commissioner: Opening Remarks Regarding Dark Pools,” October 21, 2009.

2

by matching orders within the exchange’s bid and ask. Unlike the exchange, the dark pool has no market makers through which to absorb excess order flow and thus cannot guarantee execution. Sending an order to the dark pool therefore involves a trade-off between potential price improvement and the risk of no execution. Execution risk in the dark pool drives my results. Because matching in the dark pool depends on the availability of counterparties, some orders on the “heavier” side of the market—the side with more orders—will fail to be executed. These unexecuted orders may suffer costly delays. Because informed orders are positively correlated with the value of the asset and therefore with each other, informed orders are more likely to cluster on the heavy side of the market and suffer lower execution probabilities in the dark pool. By contrast, liquidity orders are less correlated with each other and less likely to cluster on the heavy side of the market; thus, liquidity orders have higher execution probabilities in the dark pool. This difference in execution risk pushes relatively more informed traders into the exchange and relatively more uninformed traders into the dark pool. Under natural conditions, this self selection lowers the noisiness of demand and supply on the exchange and improves price discovery. The main intuition underlying my results does not hinge on the specific trading mechanisms used by a dark pool. For example, a dark pool may execute orders at the midpoint of the exchange bid and ask or operate as a nondisplayed limit order book. As I show, with both of these mechanisms, traders face a trade-off between potential price improvement and execution risk. Dark pools that operate as limit order books are, however, relatively more attractive to informed traders because limit orders can be used to gain execution priority and thus reduce execution risk. This result suggests that informed traders have even stronger incentives to trade on the exchange under a “trade-at” rule, which requires that trading venues that do not quote the best price either to route incoming orders to venues quoting the best price or to provide incoming orders with a sufficiently large price improvement over the prevailing best price. The impact of a trade-at rule on price discovery complements previous fairness-motivated arguments that displayed orders—which contribute to pre-trade transparency—should have strictly higher priority than do nondisplayed orders at the same price.2 Dark pools do not always improve price discovery. For example, in the unlikely event that liquidity traders push the net order flow far opposite of the informed traders, the
2 For example, the Joint CFTC-SEC Advisory Committee (2011) has noted: “Under current Regulation NMS routing rules, venues cannot ‘trade through’ a better price displayed on another market. Rather than route the order to the better price, however, a venue can retain and execute the order by matching the current best price even if it has not displayed a publicly accessible quote order at that price. While such a routing regime provides order execution at the current best displayed price, it does so at the expense of the limit order posting a best price which need not receive execution.”

3

on the fragmentation of order flow between an exFor example. 4 “Pinging” orders are marketable orders that seek to interact with displayed or nondisplayed liquidity. This size-discovery benefit of dark trading has been widely acknowledged by market participants and regulators. The welfare implications of dark pools could naturally depend on elements outside the setting of my model. my study additionally reveals that dark pools with more discretion in execution prices are more attractive to informed traders. The focus of this paper—i. Most other existing models of dark pools either exogenously fix the strategies of informed traders. and Wuyts (2009) and Buti. Order routing means sending orders from venue to venue. if a dark pool cannot execute an order because there is no counterparty. Finally.. see Bond. the pricediscovery effect of dark pools complements their “size discovery” function. Edmans. Ye (2011). To the best of my knowledge. and Goldstein (2012) for a survey on the literature that studies the effects of financial markets on the real economy.4 These and other procedures used by some dark pools may well contribute to concerns regarding their impact on price discovery. whereas the endogenous venue choices of liquidity traders are critical to my results.3 In addition. 3 4 . and capital formation. which I focus on in this paper. such as how price discovery and liquidity affect production decisions. and “indication of interest” (IOI). more informative orders often lead to better price discovery but also tend to worsen adverse selection on the exchange. or do not consider the role of asymmetric information regarding the asset value. typically by algorithms. as in Hendershott and Mendelson (2000). however. as in Degryse. for analytical tractability I have abstracted from some of the trading practices that are applied in dark pools. Moreover. the dark pool can route the order to another dark pool. such as “pinging. assumes exogenous choices of trading venues by liquidity traders. Ready. Rindi. this paper is the first to show that the addition of a dark pool can improve price discovery. 2010. which results in wider spreads and higher price impacts. 2012).e.presence of a dark pool can exacerbate the misleading inference regarding the asset value. and Werner (2011b). For example. Pinging is sometimes used to learn about the presence of large hidden orders. although these practices are distinct from the implications of execution risk. by which large institutional orders are executed without being revealed to the broad market. and today only a handful of dark pools execute large orders (Securities and Exchange Commission. Van Achter. Indeed. My finding stands in contrast to that of Ye (2011). who studies the venue choice of a large informed trader in the Kyle (1985) framework and concludes that the addition of a dark pool harms price discovery on the exchange. Going beyond the midpoint-matching mechanism. An IOI is an electronic message that contains selected information (such as the ticker) about an order and is sent by a trading venue (such as a dark pool or a broker) to a selected group of market participants in order to facilitate a match.” order routing. better price discovery needs not coincide with higher liquidity or welfare. asset allocation. which may further route the order into the market.

Lobo. and Viswanathan (2007). Jiang. Section 6 discusses these implications. Degryse. Rindi. Nimalendran and Ray (2012). or Reg NMS (Securities and Exchange Commission. However. and informed traders more actively participate in dark pools when asymmetric information is more severe or when the dark pool allows more discretion in execution prices. For example. informed traders and liquidity traders tend to cluster by time (Admati and Pfleiderer. O’Hara and Ye (2011). McInish. informed traders cluster less with liquidity traders in the dark pool than on the exchange because informed traders face higher execution risk in the dark pool. My results have several empirical implications. In contrast with the mechanism studied in their paper.5 A watershed event for the U. 1989. related empirical evidence documented by Ready (2012). and Werner (2011a).S. Ye (2010). Early dark pools were primarily used by institutions to trade large blocks of shares without revealing their intentions to the broad market. and van Kervel (2011). rather than intermediaries. 2 An Overview of Dark Pools This section provides an overview of dark pools. I tailor this discussion for the market structure and regulatory framework in the United States. Canada. In exchange markets. Easley. and Weaver (2011). which abolished rules that had protected the manual quotation systems of incumbent exchanges. Such predatory trading is modeled by Brunnermeier and Pedersen (2005) and Carlin.change and a dark pool—differs from the focus of prior studies on competition among multiple markets. and Asia operate similarly. and Upson (2011). Buti. dark pools had low market share. 1988) or by location (Pagano. volume correlation across stocks is higher on exchanges than in dark pools. I discuss why dark pools exist. higher volumes of dark trading lead to wider spreads and higher price impacts on exchanges. and O’Hara (1996) suggest that the purchase of retail order flows (“cream-skimming”) by regional exchanges results in higher order informativeness on the NYSE. Chowdhry and Nanda. in order to avoid being front-run. the model predicts that higher order imbalances tend to cause lower dark pool activity. For concreteness. as well as discussing recent. among others. In doing so. Dark pools in Europe. Before 2005. and what distinguishes them from each other. equity market was the adoption in 2005 of Regulation National Market System. de Jong. informed traders from liquidity traders. how they operate. as modeled here. 2005). 5 5 . to separate. at least partially. 1991). for example. Related to the effect captured by my model. Keifer. in my model dark pools rely on self selection.

The right axis plots the market shares of dark pools as a percentage of the total consolidated volume. and 13 “systematic internalizers” as of Septempber 2010. Dark pools have gained market share for reasons that go beyond recent regulations For example. see European Commission (2010). estimated by Tabb Group. A notable exception to the decline in consolidated volume occurred around the “Flash Crash” of May 2010. 16 14 12 Billion shares 12 10 8 6 4 2 0 4 2009 2010 2011 10 8 6 16 Average daily volume (left) Dark pool market share (%) Tabb Rosenblatt 14 Reg NMS encouraged newer and faster electronic trading centers to compete with the incumbents. equity trading volume and the market share of dark pools. or “dark.S. estimated by Tabb Group and Rosenblatt Securities.” venues. dark pools and broker-dealer internalization are considered opaque. 5 electronic communication networks (ECNs).Figure 1: U. As of September 2009. or “lit. The left axis plots the daily consolidated equity trading volume in the United States. as well as the market share of dark pools during the same periods. estimated by Tabb Group and Rosenblatt Securities. the market share of dark pools roughly doubled from about 6. according to CFA Institute (2009).” venues. Since Reg NMS came into effect. 129 “multilateral trading facilities” (MTFs). For more discussion of MiFID and European equity market structure. the adoption in 2007 of the Markets in Financial Instruments Directive (MiFID) similarly led to increased competition and a fast expansion of equity trading centers. According to their data. the United States had about 10 exchanges. equity markets from July 2008 to June 2011. In Europe. 32 dark pools. Exchanges and ECNs are referred to as transparent. and over 200 brokerdealers (Securities and Exchange Commission.6 Figure 1 shows the consolidated volume of U. whereas consolidated equity volume dropped persistently from about 10 billion shares per day in 2008 to about 7 billion shares per day in 2011. European equity market had 92 regulated markets (exchanges).S. a wide variety of trading centers have been established. 2010). 6 6 .5% in 2008 to about 12% in 2011.

This need has increased in recent years as the order sizes and depths on exchanges have declined dramatically (Chordia. 2011). Barclays LX. Instinet is another agency broker that operates scheduled and continuous dark pools. Examples in this group include block-crossing dark pools such as ITG Posit and Liquidnet. these dark pools are not necessarily “agency only. or “pegged” orders. We can categorize them. Because Group-1 dark pools rely on lit venues to determine execution prices. Examples include Getco and Knight. As the market moves. Further. Certain investors. Dark pools differ from each other in many ways. Finally. dark pools operate as continuous nondisplayed limit order books. they typically do not provide direct price discovery. Group-2 dark pools derive their own execution prices from the limit prices of submitted orders.” Dark pools in the third group act like fast electronic market makers that immediately accept or reject incoming orders. In this group. Dark pools operated by exchanges typically use midpoint matching as well. and UBS PIN. Price discovery can therefore take place. limit. Unlike Group-1 dark pools that execute orders at the market midpoint or VWAP. 8 7 7 . simply need nondisplayed venues to trade large blocks of shares without alarming the broad market. broker-dealers handling customer orders have strong incentives to set up their own dark pools. accepting market. Like the second See also Ready (2012) for a discussion of these two dark pools. Morgan Stanley MS Pool. the limit price of a pegged order moves accordingly. Roll. Goldman Sachs Sigma X. Dark pools in the first group match customer orders by acting as agents (as opposed to trading on their own accounts). including Credit Suisse Crossfinder. roughly. In this sense. Another difference is that Group-2 dark pools may contain proprietary order flows from the broker-dealers that operate them.7 Posit crosses orders a few times a day at scheduled clock times (up to some randomization). or the midpoint.8 This group includes many of the dark pools owned by major broker-dealers.designed to encourage competition. into the three groups shown in Table 1. Within the second group. although in recent years it has also offered continuous crossing. the offer. and Subrahmanyam. Citi Match. These derived prices include the midpoint of the national best bid and offer (NBBO) and the volume-weighted average price (VWAP). Pegged orders are limit orders with the limit price set relative to an observable market price. Liquidnet is integrated into the order-management systems of institutional investors and alerts potential counterparties when a match is found. such as the bid. dark pools attract investors by offering potential price improvements relative to the best prevailing bid and offer on exchanges. where they can better match customer orders internally and therefore save trading fees that would otherwise be paid to exchanges and other trading centers. such as institutions. transaction prices are typically derived from lit venues.

offer some price discovery and conCiti Match. Overviews of dark pools and nondisplayed liquidity are also provided by Johnson (2010). may Goldman Sachs Sigma X. and customer-to-customer Credit Suisse Crossfinder. A glossary of key model variables can be found in Appendix D. The order-book setting provides additional insights regarding the effect of the dark pool crossing mechanism for price discovery. Each trader chooses whether to trade on a transparent exchange or in a dark pool. Liquidnet. and International Organization of Securities Commissions (2011). transaction prices on these platforms are not necessarily calculated from the national best bid and offer using a transparent rule. Group-3 dark pools typically trade on their own accounts as principals (as opposed to agents or marketplaces). Butler (2007). Carrie (2008). Types Matching at exchange prices Examples ITG Posit. CSA/IIROC (2009). Section 4 models a dark pool that operates as a nondisplayed limit order book. typically execute orders at midpoint or VWAP. tain proprietary order flow Morgan Stanley MS Pool. European Commission (2010).Table 1: Dark pool classification by trading mechanisms. In contrast with dark pools in Groups 1 and 2. Securities and Exchange Commission (2010). Barclays LX. A dynamic equilibrium with sequential arrival of traders is characterized in Section 5. Most broker-dealer dark pools. Appendix A discusses additional institutional features of dark pools. Instinet Typical features Mostly owned by agency brokers and exchanges. 3 Modeling the Exchange and the Dark Pool This section presents a two-period model of trading-venue selection. 8 . typically trade as principal Nondisplayed limit order books Electronic market makers group. UBS PIN Getco and Knight High-speed systems handling immediate-or-cancel orders. The dark pool modeled in this section passively matches orders at the midpoint of the exchange’s bid and ask.

orders on the “heavier side”— the buyers’ side if buy orders exceed sell orders. A key objective of this section is to analyze price discovery. in particular P1 . that is. for the fundamental value v of the asset. Thus. is modeled in Section 4. Closing the the dark pool in period 2 is without loss of generality because once the dividend v is announced in period 2. For-profit traders and liquidity traders.3. all of the same size. exchange trading is costless. which is the expected asset value. the model of this section is not exactly the same as that of Glosten and Milgrom (1985) because orders here arrive in batches. The market maker also announces the exchange closing price P1 .1 Markets and traders There are two trading periods. The execution price of dark pool trades is the midpoint of the exchange bid and ask. The exchange is open in periods 1 and 2. The exchange here is thus similar to that modeled by Glosten and Milgrom (1985). Two trading venues operate in parallel: a lit exchange and a dark pool. then QS of the QB buy orders are randomly selected. Unmatched orders are returned to the order submitter at the end of period 1. Market orders sent to the exchange arrive simultaneously. For example. There is an infinite set of infinitesimal traders of each type. At the end of period 2. a risk-neutral market maker sets competitive bid and ask prices. and the sellers’ side if sell orders exceed buy orders—are randomly selected for matching with those on the “lighter” side. the market maker announces the volume Vb of exchange buy orders and the volume Vs of exchange sell orders. to be executed against the QS sell orders at the mid-market price. On the exchange. Sequential arrival of orders is considered in Section 5. if the dark pool receives QB buy orders and QS < QB sell orders. The closing price P1 is also the price at which the market maker is willing to execute a marginal order at the end of period 1. An alternative dark pool mechanism. arrive at the beginning of period 1. As described in Section 2. all risk-neutral. The dark pool executes (or “crosses”) orders in period 1 and is closed in period 2. also known simply as the “midpoint” or “mid-market” price. 2. exchange sell orders are executed at the bid. The asset value v is publicly revealed at the beginning of period 2. In the dark pool. Exchange buy orders are executed at the ask. 9 9 . instead of sequentially. this midpoint execution method is common in dark pools operated by agency brokers and exchanges. An order submitted to the dark pool is not observable to anyone but the order submitter.9 After period-1 orders are executed. σ > 0 is the volatility of the asset value. the informativeness of these announcements. an asset pays an uncertain dividend v that is equally likely to be +σ or −σ. equally likely. conditional on Vb and Vs . a nondisplayed limit order book. For-profit traders As I describe shortly. denoted by t = 1.

d. That is. Similarly. + Note that the distribution of depends on n. A similar construction applies for the total liquidity sell orders Z − . Z + can be constructed as the sum of n independently and identically + distributed random variables {Zin }. at a cost. z + Similarly. 1]. Indeed. Thus. Liquidity buyers and liquidity sellers arrive at the market separately (not as netted). given this liquidity trader’s order size. For-profit traders who do not acquire the information do not trade.11 In particular. converges in n to the prior distribution of Z − . By the independence of Zin and {Zjn }j=i . as n → ∞. and infinitely divisible. we have + E(Zin ) + + P(Zin < z) = 1 − P(Zin ≥ z) ≥ 1 − → 1. and can be viewed as the order sizes of n liquidity buyers and n liquidity sellers. is the same as the unconditional joint distribution of Z + and Z − . the conditional joint distribution of Z + and Z − .12 I denote by 0. 12 We denote by Φ the probability distribution of Z + and show that. After observing v. converges to the prior distribution of Z + + + as n → ∞. n 10 Z+ ∼ i=1 + Zin + Zin . ∞) → [0. informed traders submit buy orders (in either venue) if v = +σ and submit sell orders if v = −σ. I let µI be the mass of informed traders. In this setting. for each integer n. this amounts to showing that Zin converges to zero in + distribution. for any z > 0. Φ(z | Zin ) → Φ(z) j=1 Zjn . As long as per-capita trading size is finite. The proof 10 . with the cumulative distribution function F : [0. using Markov’s inequality and the fact that E(Zin ) converges to zero as n → ∞.5µz Trading one unit per capita is without loss of generality because each informed trader is infinitesimal and has zero mass. an informed trader’s order still has zero mass. perfect information about v. 11 More specifically. given Zin . the total liquidity buy orders Z + can be viewed as the aggregate demand by n liquidity buyers. for all z > 0. the conditional distribution of n + + + + as n → ∞. we can interpret a market with infinitely many liquidity traders as the “limiting case” of a market with n liquidity buyers and n liquidity sellers as n → ∞. their signed trading interest is therefore Y = sign(v) · µI . each liquidity trader’s order size has zero mean and zero variance. and the qualitative nature of equilibria does not change.i. These information-acquisition costs are distributed across for-profit traders. and liquidity buyers and sellers become infinitesimal. and thus become informed traders. whose order sizes are independently and identically distributed random variables. for each integer n. ∞) with positive density functions. in the limit. the conditional distribution of Z − . given Zin . the mean and variance of Zin and Zin converge to zero. The mass Z + of liquidity buy orders and the mass Z − of liquidity sell orders are non-negative.have a mass of µ > 0. for each i. and can potentially trade one unit of the asset per capita. Infinite divisibility means that. independent and identically distributed on [0.10 For¯ profit traders can acquire. there exist n − i. As + − n → ∞. because. I assume that the variance of Zin is finite. That is. random variables Zin such that n Z− ∼ i=1 + {Zin } − {Zin } − Zin .

Informed and liquidity traders cannot post limit orders on the exchange. 0) | v > 0] = γi σ (1) per unit of undesirable asset position. These funding costs {γi } are distributed across liquidity traders. they can trade only with the exchange market maker or by sending orders to the dark pool. a liquidity buyer who is already short one unit of the asset has a loss of σ if v = σ. and their probability distributions are common knowledge. with a twice-differentiable cumulative distribution function G : [0. of her undesired position. for some Γ ∈ (1. random variables v. the minimum collateral requirement per unit asset held is the expected loss. liquidity buyer i thus incurs a delay cost of ci = γi E[max(v. 1]. with the exception that informed traders observe v. Failing to trade in period 1. each unit of collateral has a funding cost of γi per period. The key is that liquidity traders differ in their desires for immediacy. and the costs of information-acquisition and delay are all independent. conditional on a loss. For each liquidity trader. Γ) → [0. We could alternatively interpret this delay cost as stemming from risk aversion or illiquidity. and a gain of σ if v = −σ. For trader i. Figure 2 illustrates the sequence of actions in the two-period model. The collateral requirement in this case is σ. For example.5σz the variance of Z + (and Z − ). 11 . Liquidity traders must hold collateral to support their undesired risky positions. Z + and Z − are unobservable. Realizations of Y .Figure 2: Time line of the two-period model. captured by the delay cost ci = γi σ. A delay in trade is therefore costly. for a liquidity seller’s inference is symmetric. The delay costs of informed traders. and hence know Y . Z − . Finally. | Period 1 Exchange announces closing price Orders executed Period 2 | Value v announced Remaining orders executed on exchange Value v paid Traders select venue or delay trade 2 the mean of Z + (and Z − ) and by 0. by contrast. Z + . A like delay cost applies to liquidity sellers. ∞]. stem from the loss of profitable trading opportunities after v is revealed in period 2.

I refer to S as the “exchange spread. the net profit of an informed trader is σ − S. Specifically. αd . Because the market maker breaks even in expectation. in period 1. I first derive the equilibrium exchange bid and ask. I let αe and αd be candidates for the equilibrium fractions of liquidity traders who. The remaining fraction 1 − β of informed traders trade on the exchange. Because of symmetry and the fact that the unconditional mean of v is zero. and the trading strategies of informed and liquidity traders. where S is the exchange’s effective spread. and the exchange bid is −S. all traders who have not traded in period 1—including those who deferred trading and those who failed to execute their orders in the dark pool—trade with the market maker at the unique period-2 equilibrium price of v. choose not to submit orders in period 1 and delay trade to period 2. respectively. the competitive market maker breaks even in expectation. and all traders maximize their expected net profits. informed traders never delay their trades as they will have lost their informational advantage by period 2. αd . Next. the market participation strategies of for-profit traders. αe ).” As in Glosten and Milgrom (1985). Given the participation fractions (β. assuming equilibrium participation fractions (β. the midpoint of the market maker’s bid and ask is zero. The information-acquisition cost of the marginal for-profit trader.2 Equilibrium An equilibrium consists of the quoting strategy of the exchange market maker.3. the mass of informed traders on the exchange is (1 − β)µI . We let β be the period-1 fraction of informed traders who send orders to the dark pool. we have that 0 = −(1 − β)µI (σ − S) + αe µz S. For simplicity. (1 − β)µI + αe µz (3) The dark pool crosses orders at the mid-market price of zero. the absolute difference between the exchange transaction price and the midpoint. and the expected mass of liquidity traders on the exchange is αe E(Z + + Z − ) = αe µz . The remainder. (2) which implies that S= (1 − β)µI σ. Given the value σ of information and the exchange spread S. who is indifferent 12 . the exchange ask is some S > 0. Therefore. send orders to the exchange and to the dark pool. αe ). the exchange bid and ask are set before exchange orders arrive. (Obviously. In equilibrium.) Once the asset value v is revealed in period 2. I derive the equilibrium mass µI of informed traders. α0 = 1 − αe − αd .

If. Because informed buyers trade in the same direction. µ). do not observe v. we have 1 > r+ > r− > 0. (7) αd Z − + βµI αd Z + . on the other hand. then any dark pool 13 . (4) has a unique solution µI ∈ (0. Liquidity buyers. (1 − β)µI + αe µz (4) For any fixed β ≥ 0 and αe > 0. then liquidity buyers have the crossing probability r+ = E min 1. I turn to the equilibrium trading strategies. by the exact law of large numbers (Sun. I calculate the expected payoffs of an informed buyer and a liquidity buyer.between paying for information or not. Without loss of generality. In other words. 2006). (5) where the denominator and the numerator in the fraction above are the masses of buyers and sellers in the dark pool. αd Z − αd Z + + βµI . their dark pool crossing probability (r+ + r− )/2 is greater than informed traders’ crossing probability r− . If the dark pool contains only liquidity orders (that is. they have the dark pool crossing probability of r− = E min 1. The equilibrium is then naturally determined by conditions characterizing marginal traders who are indifferent between trading on the exchange and in the dark pool. I focus on the strategies of buyers. (6) Because liquidity buyers assign equal probabilities to the two events {v = +σ} and {v = −σ}. Suppose that αd > 0. correlated informed orders have a lower execution probability in the dark pool than relatively uncorrelated liquidity orders. In the main solution step. We thus have ¯ µI = µF (σ − S) = µF ¯ ¯ α e µz σ . β = 0). If informed traders are buyers. ¯ Finally. is also σ − S. respectively. the mass of informed traders in equilibrium is µF (σ − S). informed traders are sellers. Obviously. however. on the exchange and in the dark pool. for all β > 0. Because all for-profit traders with lower information-acquisition costs become informed. then liquidity buyers have the crossing probability r− in the dark pool.

crossing in the dark pool implies a positive adverse selection cost because execution is more likely if a liquidity trader is on the side of the market opposite to that of informed traders. In particular. respectively. respectively. Xd (c) = − r+ − r− r+ + r− σ−c 1− 2 2 . (14) That is. Wd − Xd (c) = r+ + r− r+ + r− σ+c 1− 2 2 ≤ σ = We − Xe . the dark pool is more attractive to liquidity traders than to informed traders. r measures the degree to which liquidity orders are balanced. trading on the exchange. X0 (c) = −c. each trader’s optimal venue choice is a corner solution with probability one. ¯ The expected profits of an informed buyer on the exchange and in the dark pool are. For αd = 0. Xe = −S. and trading in the dark pool are. ¯ Z− Z+ . (9) (10) I denote by c the delay cost of a generic liquidity buyer per unit of asset position. in the remaining of the paper the net profits and delay costs of liquidity traders refer to profits and costs per unit of asset. provided c ≤ σ. unless otherwise specified. Per¯ fectly balanced liquidity orders correspond to r = 1.buy order has the execution probability r = E min 1. Wd = r− σ. by risk neutrality. (8) For our purposes. (11) (12) (13) The terms on the right-hand side of (13) are the liquidity trader’s adverse selection cost and delay cost in the dark pool. this adverse-selection cost is zero. For β = 0. (14) implies that a liquidity trader with a delay cost of σ (or a funding cost of γ = 1) behaves in the same way as 14 . It is without loss of generality to focus on the venue decision for one unit of asset because. We = σ − S. We − Xe = σ. This buyer’s per-unit net payoffs of deferring trade. respectively. For all delay cost c ≤ σ. For simplicity. I define r+ = r− = 0. For β > 0. From (9) and (12). relative to the exchange.

(1 − β)µI + (1 − G(1))µz r+ − r− αd = G(1) − G + . αd ∈ (0. β = 0). In this case. µI + (1 − αd )µz (1 − αd )µz µI = µF ¯ σ . In addition. αd = αd . ¯ ∗ where β ∗ . The equilibrium is then determined by the marginal liquidity trader who is indifferent between trading on the exchange and trading in the dark pool. Xd (c) − X0 (c) = − r+ − r− r+ + r− σ+ c. a liquidity trader with a delay cost of σ is also indifferent between the two venues. First. as well as by the marginal for-profit trader who is indifferent about whether to acquire the information. In this case. αd = αd . 2 2 (15) So a liquidity trader with a funding cost of γ = (r+ − r− )/(r+ + r− ) is indifferent between deferring trade and trading in the dark pool. There exists a unique threshold volatility σ > 0 such that: ¯ ∗ ∗ 1. The second indifference condition (19) then follows 15 . there exists an equilibrium (β = β ∗ . Thus. the exchange spread is low.e. If the volatility σ is sufficiently low. then there exists an equilibrium (β = 0. 1). If σ ≤ σ . informed traders joint liquidity traders in the dark pool to avoid the higher exchange spread. thus. By (14). as shown in (18). the price-improvement benefit of the dark pool is lower than the cost of execution risk. r + r− (1 − G(1))µz µI = µF ¯ σ . G(1)]. If and only if σ > σ . αe = 1 − αd ). informed traders avoid the dark pool (i. αe = 1 − G(1)). the equilibrium is determined by three indifference conditions. If the volatility σ is sufficiently high. β ∈ (0. α0 + αd = G(1) and αe = 1 − G(1). G(1)] and µ∗ solve I G−1 (αd )(1 − r) = ¯ µI . Proposition 1. and µ∗ solve I r− = 1 − (1 − β)µI . but we outline its main intuition here.an informed trader. (1 − β)µI + (1 − G(1))µz (18) (19) (20) The proof of Proposition 1 is provided in Appendix C. informed traders must be indifferent between trading in either venue. µI + (1 − αd )µz (16) (17) ∗ 2. Thus. where ¯ ∗ αd ∈ (0.

For example. as αd varies in the neighborhood of αd . 1) solve I µI = G−1 (1 − αe ) µI + α e µz α e µz µI = µF ¯ σ . 14 ∗ Selecting the stable equilibrium corresponding to the smallest αd is arbitrary but without loss of generality. which allows me to compute the comparative statics of the selected equilibria. This exchange-only equilibrium. Thus. both sides of (16) strictly increase in αd . the left-hand side of (16) crosses the right-hand side from below. multiple equilibria may arise. µI and β are uniquely determined. if there is an equilibrium in which. αd is “pushed ∗ back” to αd and the equilibrium is restored. Here. With a linear G. may also be interpreted as one in which a dark pool is open but no trader uses it. once αd is determined in equilibrium. Thus. stated below. for example. There is a symmetric argument for a small downward perturbation to ∗ αd − . and both sides of (21) strictly decrease in αe . the condition (63) is also necessary for the existence of solutions to (16)-(17). µI + α e µz Equilibrium selection The equilibria characterized in Proposition 1 need not be unique among all equilibria solving (16)-(17) and (18)-(20).13 I use stability as an equilibrium selection criterion. this equilibrium would not be stable to local perturbations. 15 ∗ ∗ If. we can characterize an equilibrium for a market structure in which only the exchange is operating and the dark pool is absent. the fraction α0 of liquidity traders who delay trade must be strictly positive because informed traders introduce adverse selection into the dark pool. Among the equilibria characterized by ∗ Case 1 of Proposition 1. As long as the selected equilibrium is stable. then the marginal liquidity trader has a higher cost in the dark pool than on the exchange. this equilibrium exists and is robust to small perturbations. both sides of (19) strictly increase in αd . given the absence of a single-crossing property. and therefore migrates out of the dark pool. as shown in the proof of Proposition 1. I select that with the smallest liquidity participation αd in the ∗ dark pool among those with the property that. αd is perturbed to αd + for sufficiently small > 0. too. comparative statics calculated later follow through. With only an exchange and no dark pool. there exists an equilibrium in ∗ ∗ which β ∗ = αd = 0. Similarly. the left-hand side of (16) crosses the right-hand side from above.14 Under the conditions of Proposition 1. Similarly. The third condition (20) says that the marginal for-profit trader is indifferent about whether to acquire the information. One special condition that guarantees the uniqueness of the equilibrium in Case 1 of Proposition 1 is that the distribution function G of delay costs is linear.from (15).15 Moreover. under the condition (63). and µ∗ and αe ∈ (1 − G(1). Corollary 1. as αd varies. 13 (21) (22) 16 . By contrast.

µI . all of µI . µI . 2. whereas αe is strictly decreasing in σ. as αd varies in the neighborhood of αd . adding a dark pool strictly reduces the exchange participation rate αe ¯ of liquidity traders and the total mass µI of informed traders. The exchange participation rate αe = 1 − αd of liquidity traders is strictly decreasing in σ. In a market without a dark pool (Corollary 1). whereas αd and ¯ − r are strictly decreasing in σ. r+ . Given a fixed value σ of private information. I select ∗ the equilibrium with the largest liquidity participation αe on the exchange among those ∗ with the property that. Proof.Similarly. the total mass ¯ µI of informed traders. For σ ≤ σ . αd . these selected equilibria exist and are stable. For σ ≤ σ . the left-hand side of (21) crosses the right-hand side from below. Proposition 3. I select the ∗ one with the smallest liquidity participation αd in the dark pool among those with the ∗ property that. Adding a dark pool strictly increases the exchange spread S and the total participation rate αe + αd of liquidity traders in either venue. and r− are continuous and differentiable in σ. By the argument given for Case 1 of Proposition 1. Moreover.3 Market characteristics and comparative statics I now investigate properties of the equilibria characterized by Proposition 1. αd . how does adding a dark pool affect market behavior? Proposition 2. the left-hand side of (19) crosses the right-hand side from below. µI and S/σ are strictly increasing in σ. the dark pool participation rate αd of liquidity traders. as αe varies in the neighborhood of αe . In the equilibria of Proposition 1 and Corollary 1: 1. and S are continuous and differentiable in σ. αe . and the scaled exchange spread S/σ are strictly increasing in σ. and S are continuous and differentiable in σ. Moreover. 17 . Proposition 2 and Proposition 3 below aim to answer two questions: 1. and S/σ are strictly increasing in σ. In the equilibrium of Corollary 1. For σ > σ . 3. In the equilibrium of Proposition 1: 1. among equilibria characterized by Case 2 of Proposition 1. β. µI . r+ . S. βµI . Moreover. how do market characteristics vary with the value σ of private information? 2. In a market with a dark pool and an exchange. See Appendix C.

For sufficiently high β. Finally. The right-hand side plot of Figure 3 shows the scaled exchange spread S/σ. For σ > σ . but ¯ none of the informed traders. and these additional informed traders prefer to trade in the dark pool.2. For σ ≤ σ . For σ > σ . For a low β. the scaled exchange spread S/σ increases in σ. For σ > σ .1 Participation rates and exchange spread (24) The left-hand side plot of Figure 3 shows the equilibrium participation rates in the exchange and the dark pool. This higher value of information attracts additional informed traders. We observe that informed dark pool partic¯ ipation rate β first increases in volatility σ and then decreases. these additional informed traders send orders to the exchange.3. (23) µI + (1 − G(1 − r− ))µz It is sufficient (but not necessary) for (23) that G (γ) ≤ 0 for all 1 − r ≤ γ ≤ 1 and F (c) → 1 for all c > 0. liquidity traders with low delay costs migrate out of the dark pool. See Appendix C. informed traders use both venues. because informed participation in the dark pool introduces adverse selection. in the equilibrium of Proposition 1. for some > 0. For a small value of information. specifically if σ ≤ σ . ¯ Proof. adding a dark pool raises S/σ by diverting some liquidity traders. reducing β. Thus. leading to a decline in their dark pool participation rate αd . informed orders cluster on one side of the dark pool and significantly reduce their execution probability. Consider an increase in the value of information from σ to σ + . adding a dark pool in ¯ 18 . µI r− < 1 − . We now discuss the intuition and implications of Proposition 2 and Proposition 3 through numerical examples. An increase in σ widens the exchange spread. encouraging more liquidity traders to migrate into the dark pool. Nonetheless. ¯ informed traders trade exclusively on the exchange because the exchange spread is smaller than the cost of execution risk in the dark pool. Because a higher value σ of information encourages more for-profit traders to become informed. raising β. whether a dark pool is present or not. off the exchange. The intuition for this non-monotonicity is as follows. however. the dark pool execution risk stays relatively low. adding a dark pool strictly reduces αe . the total quantity βµI of informed traders in the dark pool is strictly increasing in σ. Moreover. 3. adding a dark ¯ pool strictly increases the exchange spread S if and only if.

The right-hand side plot shows the scaled exchange spread S/σ. 2]. σz = 60. its conditional distribution after period-1 trading is completely determined by its conditional expectation E[v | P1 . all period-1 public information that is relevant for the asset value v is conveyed by the closing price P1 .05 −2 0 2 4 Lit only Lit + Dark log(σ) log(σ) 0 −2 0 2 4 this example also increases the scaled spread S/σ because the dark pool diverts more liquidity traders than informed traders. (26) That is. Vs ).6 S/σ 0. Vs ] = E[E[v | Vb . ∞). 1 0.25 0. Z + and Z − have Gamma(30. Since the market maker observes the volume (Vb .4 0. the vertical dotted line indicates the threshold volatility σ at which the equilibrium of Proposition 1 changes from Case 1 to Case 2. µ = 20. by which I mean the extent to which the period-1 announcements (P1 . the “closer” is P1 to v.3 0. Vb . Vb . −s/2 and F (s) = 1 − e for s ∈ [0. the closing price P1 is P1 = E[v | Vb . (25) Because v is binomially distributed. Vs ) are informative of the fundamental asset value v. As we will make precise shortly. 3. 1) distributions.1 0.2 0 β αd Participation rates αe 0.35 0.2 0. αd . G(s) = s/2 for s ∈ [0. The left-hand side plot shows the equilibrium participation rates (β. Vs ]. 19 . Vs ] | P1 ] = P1 . the better is price discovery.8 0.2 Price discovery Now I turn to price discovery.Figure 3: Participation rates and exchange spread. αe ) in a market with a dark pool. In both plots.15 0. Model parameters: ¯ √ ¯ µz = 60.3.

I condition on v = +σ and consider price discovery to be unambiguously “improved” if the probability distribution of R1 is “increased. Fix a small δ > 0 such that m = µz /δ is an integer. we can always take an example in which. the 2 2 density φ(·) is approximated by N ormal(0. 2 2 αe σz (30) which is the counterpart of R1 under the normal distribution. Vs ) φ Z+ = 1 [V αe b 1 V αe b + − (1 − β)µI ] · φ Z − = · φ Z− = − 1 [V αe s 1 V αe s .d. we can approximate R1 by normal R1 = 2(1 − β)µI (Vb − Vs ). P1 . − (1 − β)µI ] (27) where φ is the probability density function of Z and Z . a non-trader assigns the probability Q1 ≡ P(v = +σ | Vb . Without loss of generality. by the central limit theorem. the central limit theorem implies that the distribution of Z + is approximately normal 2 when m is large.5.17 Given v = +σ.5σz ) when µz and σz are sufficiently large. Vs ) R1 = log = log P(v = −σ | Vb . when µz and σz are large.i. 16 20 . Further. By infinite m + + divisibility. Z + can be represented as the sum i=1 Zim . and Q1 . Fixing δ. I have used the central limit theorem and the fact that φ(·) and the normal density are positive and continuous in [0. P1 is uniquely determined by the log likelihood ratio φ Z+ = P(v = +σ | Vb . Complete revelation of v = +σ corresponds to R1 = ∞ almost surely. where {Zim } are i. so R1 is has a distribution We can show this approximation as follows. the market maker sets the period-1 closing price P1 = e R1 − 1 σ. We have also used the fact that the prior distribution P(v = +σ) = P(v = −σ) = 0. Vs ) = 1 eR1 = eR1 + 1 2 P1 +1 σ (29) that the asset value is high. αe σz ). that is. Z + and Z − can be expressed as the sums of i.5µz . In general.16 Substituting into (27) the normal density function. 0. random variables. However. random variables with mean δ and 2 variance δσz /µz . Given R1 . 17 In the calculation of (30).i. since the distribution of Z + and Z − is infinitely divisible. we need to know the functional form of the density φ(·) in order to explicitly calculate R1 . eR1 + 1 (28) Conditional on P1 .” in the sense of first-order stochastic dominance.d. ∞). Vb − Vs 2 2 has a distribution close to that of N ormal((1 − β)µI .Clearly.

With a dark pool. adding a dark pool may increase the cumulative distribution of R1 . such misleading events are rare. The value σ of information is set to be the threshold value σ . for most realizations of Z. 18 21 .18 Nonetheless. under normal approximation. I(β. the conditional distribution of R1 has a higher mean. αe ) ≡ (1 − β)µI αe σz (32) is the “signal-to-noise” ratio. under normal approximation. The right-hand plot of Figure 5 shows how Q1 depends on the imbalance Z = Z + −Z − of liquidity order flow. and the We can analytically show that the expectation E[Q1 ] under normal approximation is increasing in the signalto-noise ratio I(β. The price-discovery effect of the dark pool is further illustrated in Figure 5. αe ). adding a dark pool strictly increases the scaled spread S/σ and hence the signal-to-noise ratio I(β. αe ) is increasing in the scaled exchange spread S/σ. thus harming price discovery. Again.close to that of N ormal 2 (1 − β)µI αe σ z 2 . The left-hand plot of Figure 5 shows the probability density function of Q1 . adding the dark pool increases Q1 . when the realization of R1 is sufficiently low. adding a dark pool shifts the probability density function of Q1 to the right. Nonetheless. adding the dark pool reduces Q1 . the dark pool increases the probability of extremely low realizations of Q1 . 4I(β. thus harming price discovery. Figure 4 plots the distribution function of R1 . adding a dark pool decreases the cumulative distribution of R1 and leads to a more precise inference of v. which is the mass of informed orders on the exchange (“signal”) divided by the standard deviation of the imbalance of liquidity orders on the exchange (“noise”).4 (1 − β)µI αe σz 2 ∼ N ormal 2I(β. For unlikely low realizations of Z. when the trading interests of liquidity traders are sufficiently large and opposite in direction to the informed. and on average. That is. αe ). Because liquidity trading interests are balanced in expectation. improving price discovery. ¯ so that β = 0 in the equilibria with a dark pool as well as the equilibria without a dark pool. (31) where I(β. αe )2 . but also a higher variance. with and without a dark pool. with and without a dark pool. αe )2 . adding the dark pool can exacerbate the “misleading” inference regarding the asset value. improving price discovery on average. As in Figure 4. harming price discovery in these unlikely events. For most realizations of R1 . Naturally. By Proposition 3.

3 0. Because trading is frequent and fast (with the exception that large orders can take days to fill). dark pools are most likely to concentrate shortterm information.6 0. onto the exchange. with and without a dark pool. The right-hand plot shows how Q1 depends on the order imbalance Z = Z + − Z − of liquidity order flow.4 0.8 0.2 0.2 −5 0 R1 5 −5 0 R1 5 Figure 5: The left-hand plot shows the probability density function of Q1 .4 0. under normal approximation. σ Lit only Lit + Dark Probability density 0.4 Q 0.8 −20 0 20 Z 40 dark pool is normally beneficial for price discovery.2 0.Figure 4: Distribution functions of R1 .5 1 0. when both short-term investors and long-term investors are present. Model parameters are those of Figure 4. Short-term information can be fundamental (such as merger announcements. We observe that the practical interpretation of price discovery depends largely on the horizon of information. 1 2. Moreover.2 0 −40 0.8 2 Q1 1. with and without a dark pool. Under this 22 .4 0. rather than long-term information.1 Cumulative distribution 0. The true dividend is the threshold value +¯ and other parameters are those of Figure 3.5 Probability density 0.5 Lit only Lit + Dark 0.6 1 Lit only Lit + Dark 0. under normal approximation.6 0. or macroeconomic news) or technical (such as the order flows of large institutions). earnings reports. it is natural to interpret the former as informed and the latter as uninformed.

When calculating the exchange volume. by ¯ ¯ Proposition 2. the dark pool market share can decrease in volatility σ for sufficiently large σ. because they will not be able to trade profitably. is increasing in the volatility σ. I also include the transactions of liquidity traders in period 2. informed traders who have not yet traded leave the market. µz + µI dσ (µz + µI )2 dσ where the inequality follows from the fact that limσ→0 dαd /dσ > 0 (shown in the proof of Proposition 2). In particular. therefore.3 Dark pool market share I now calculate the dark pool market share. because the total volume V = µz + µI ¯ is increasing in σ. the dark pool participation rate αd of liquidity traders and the dark pool market share Vd /V converge to zero. the exchange ¯ volume Ve can increase substantially. σ ] and (¯ . By Proposition 2. but the dark pool volume Vd may only increase mildly or even decline. 2 − (33) r+ + r− 2 + α0 µz .3. (34) (35) Ve = (1 − β)µI + αe µz + αd µz 1 − V = Ve + Vd = µz + µI (1 − β + βr− ). i. Vd = αd µz r. if the volatility σ is sufficiently low. as the volatility σ increases beyond σ . That is. the proportion of trading volume handled by the dark pool.interpretation. ¯ σ For σ ≤ σ . The market share of the dark pool is a direct empirical measure of dark pool activity. For σ ≤ σ . 3. ¯ d d(Vd /V ) = dσ dσ α d µz r ¯ µz + µI = µz r ¯ dαd α d µz r ¯ dµI · − · > 0. respectively. the expected trading volumes in the dark pool. Figure 6 further suggests that. r+ + r− Vd = βµI r + αd µz . as illustrated in the left-hand plot of Figure 6. creating a hump-shaped relation between volatility and the dark pool market share. the dark pool market share is increasing in the total volume. relative to the exchange. on the exchange. then the dark pool market share Vd /V is increasing in σ. the dark pool volume. For a sufficiently small σ < σ . as σ → 0. Thus. these volumes are differentiable in the volatility σ in each of the two intervals [0.e. I assume that once the dividend v is announced in period 2. The model also generates a similar relation between the scaled 23 . and in both venues are. ∞). Thus. the results of this paper suggest that dark pools are more attractive to long-term investors than to short-term investors.

The left-hand plot shows the volume in the two venues and the market share of the dark pool. MiFID uses a decentralized best-execution rule.3 0.Figure 6: Expected trading volume on the exchange and in the dark pool. where execution prices depend on submitted limit orders. such price discretion is often limited by “best-execution” regulations.15 0.35 0.25 0. by which investment firms decide whether an execution works for the best interest of investors.1 0.19 For example.50]. In the United States.25 0.2 20 0. I model a dark pool that operates as a nondisplayed limit order book. Although limit-order dark pools may execute orders at prices other than the midpoint. ECN. In this section. if the current best bid is $10 and the best ask is $10.2 0. Aside from confirming the basic intuition of Section 3.3 0.3 log(σ) spread S/σ and the dark pool market share Vd /V . as described in Section 2.2 0.15 0.50.05 S/σ Dark pool market share 0. as shown on the right-hand plot of Figure 6.1 0 Lit volume Dark volume −2 0 2 4 10 0 50 40 30 0. then the transaction price in any market center must be in the interval [$10. The vertical dotted line corresponds to the threshold volatility σ . also known as the “trade through” rule. In Europe. stipulates that transaction prices in any market center—including dark pools. $10. 4 Dark Pools as Nondisplayed Limit Order Books So far we have studied a dark pool that crosses orders at the midpoint of the exchange bid and ask. the Order Protection Rule. Parameters are those of Figure 3.4 0.1 0.05 0. 19 24 . More recently. and broker-dealer internalization—cannot be strictly worse than the prevailing national best bid and offer (NBBO). The right-hand plot shows the dark pool market share as a function of the scaled spread S/σ. this section offers additional insights regarding the impact of dark pool mechanisms on the participation incentives of informed traders.4 0. ¯ 60 Dark/Total (left) 0. regulators have also proposed a stricter “trade-at” rule.

Proposition 4.g. The dark-pool execution price. supply) at p∗ . there exists an equilibrium (β = 0. Moreover. execution prices in dark pools must be strictly better than the best bid or offer on all displayed venues.Under a trade-at rule. Boulatov and George (2010) model a nondisplayed market in which informed traders submit demand schedules (i. the model of this section is identical to that of Section 3. The dark pool executes orders by price priority.5 cent] for most securities) to the quoted best bid or offer. and I model its trading mechanism as a double auction. demand) at p∗ is equal to the aggregate sell limit orders (i. conditional on order execution at each possible price in the interval [−xS. The trade-through rule currently applied in the United States corresponds to x = 1. σ µI + (1 − αd )µz (1 − αd )µz µI = µF ¯ σ . 20 25 . If c ∈ [0. αe = 1 − αd ). limit orders). 50 mils [0. where S > 0 is the exchange spread and x ∈ [0. is determined such that the aggregate limit buy orders (i. indicating a mandatory price improvement of zero. G(1)] and µ∗ solve I G−1 (αd ) − µI xS · (1 − rx ) = ¯ .” I now describe and solve a simple model of a limit-order dark pool that operates under a trade-at rule.e. I model the effect of a trade-at rule by assuming that transaction prices in the dark pool must be within the interval [−xS. a liquidity buyer (resp. µI + (1 − αd )µz (36) (37) In this equilibrium with a fixed x: 1. Proposition 4 below characterizes an equilibrium that is analogous to Case 1 of Proposition 1. where ¯ ∗ αd ∈ (0. I have not characterized an equilibrium in which some informed traders send orders to the limit-order dark pool. there exists a unique threshold volatility σ (x) > 0 with the property ¯ ∗ ∗ that. endogenous venue selection of liquidity traders is a key modeling objective of this paper. seller) with a delay cost of c quotes a limit For tractability reasons.e. including exchanges. 1] captures the strictness of the trade-at rule. for any σ ≤ σ (x). p∗ . indicating a price improvement of the entire effective spread S. A midpoint-matching mechanism corresponds to x = 0.e. xS].20 This result sheds light on how the trade-at rule affects the dark pool participation of informed traders. By contrast. the Joint CFTC-SEC Advisory Committee (2011) recommends that the SEC consider “its rule proposal requiring that internalized or preferenced orders only be executed at a price materially superior (e. αd = αd . not only the midpoint. xS]. For example. With the exception of this trade-at rule. In a market with an exchange and a dark pool that implements a double auction. A modeling challenge with informed participation in the limit-order dark pool is to calculate the expected loss of liquidity traders. Their model is tractable partly because their uninformed traders are noise traders and hence do not internalize the costs of trading against informed traders. xS).

2. 3. If multiple equilibria exist. G−1 (αd )σ]. seller) with a delay cost of c quotes a limit price of xS (resp. The double auction here has the institutional restriction that transaction prices are bounded by the trade-at rule. If c ∈ [xS. I select the ∗ equilibrium with the lowest αd among those with the property that. so she submit a dark pool buy order with the limit price c. the trade-at rule becomes binding at the price xS.21 If her decay cost c < xS. ¯ This strategy is reminiscent of the truth-telling strategy of MacAfee (1992). The equilibrium is determined by the marginal liquidity trader who is indifferent between the two venues. the left-hand side of (36) crosses the right-hand side from below. When the delay cost c is sufficiently high. In equilibrium. the volatility threshold σ (x) is strictly decreasing in x. Moreover. The expressions of σ (x) and p∗ in equilibrium are provided in Appendix C. ¯ Proof. the mass µI of informed traders. in equilibrium liquidity traders who have higher delay costs submit more aggressive orders (i. Moreover. Liquidity traders with delay costs higher than G−1 (αd )σ trade on the exchange. to submit a buy order whose limit price is equal to her delay cost. The equilibrium of Proposition 4 with a limit-order dark pool is qualitatively similar to the equilibrium characterized in Case 1 of Proposition 1. then a liquidity buyer (resp. because a liquidity buyer’s order is infinitesimal and has zero impact on the execution price p∗ . Because σ (x) is decreasing in x. −xS) in the ∗ dark pool. and the scaled exchange spread S/σ are all strictly increasing in the value σ of information. so the liquidity buyer selects the highest limit price allowed. and the marginal for-profit trader who is indifferent about whether to acquire the information. xS. The intuition for a liquidity seller is symmetric.∗ price of c (resp. a strictly positive mass of liquidity buyers set the limit price xS and are rationed with a positive probability. shown in (37). −c) in the dark pool. buy orders with higher limit prices and sell orders with lower limit prices). The dark pool execution price is given by (79) in the appendix.e. the liquidity buyer trades on the exchange in order to avoid the risk of being rationed at the price xS. 21 26 . Proposition 4 further reveals that the trade-at rule has a material effect for the participation of informed traders in the dark pool. shown in (36). who considers a double auction with finitely many buyers and sellers. ¯ Naturally. See Appendix C. that is. The dark pool participation rate αd of liquidity traders. If c ≥ xS. 1). for x ∈ (0. she wishes to use a “truth-telling” strategy. as αd varies in a ∗ neighborhood of αd . the trade-at rule is not binding.

Boulatov and George (2010) model how informed 27 . the less attractive is the dark pool to informed traders. With midpoint crossing (x = 0). and Werner (2011). about 95% of liquidity traders in the dark pool set the most aggressive limit price. The right-hand plot of Figure 7 shows that the volatility threshold σ (x) is strictly decreasing in x. Since their model allows traders to post limit orders on the displayed market. so the dark pool provides a price improvement equal to 20% of the exchange spread S. For example. It also predicts that dark pools operating as limit order books are more likely to attract informed traders and impatient liquidity traders than dark pools crossing at the midpoint. In their model.8. The lower is x. Although she would be rationed at the price xS. On the other hand. In this example. which suggest that displayed orders should have strictly higher priority than nondisplayed orders at the same price (Joint CFTC-SEC Advisory Committee. Rindi. The dark pool transaction price in this case is about 0. ±xS. 2011). Under the current trade-through rule (x = 1). A desirable (and nontrivial) extension of my price-discovery model is to fully allow limit orders in both the exchange and the dark pool. The extensive literature on displayed limit order books is surveyed by Parlour and Seppi (2008). In this example. my model focuses on asymmetric information and price discovery.the stricter is the trade-at rule. who study the effect of tick size for market quality. The left-hand plot of Figure 7 shows the dark pool orders in the equilibrium of Proposition 4. which are not offered in my model. she would select the most aggressive permissible limit price. and this extension is left for future research. in order to maximize her execution probability.007. If an informed buyer were to deviate to the dark pool. she would only compete with those liquidity traders who have a delay cost of xS or higher. The effect of the trade-at rule on informed participation in dark pools complements prior fairness-motivated arguments. Wen.22. this volatility threshold is reduced to about 0. xS. x = 0. Proposition 4 predicts that implementing a trade-at rule is likely to reduce informed participation in dark pools. In particular. My model of a limit-order dark pool is related to and complements that of Buti. The intuition is as follows.35. which are absent in their model. ¯ informed traders avoid the dark pool if the value σ of information is lower than about 0. the less scope there is for the informed trader to “step ahead of the queue” and gain execution priority. a midpoint dark pool with x = 0 has the greatest effectiveness in discouraging informed traders to participate. The model of this section also differs from existing studies of nondisplayed markets that operate alone. a limit order book with a subpenny tick size is similar to a dark pool studied in this section. it generates predictions on quote depths.

4 0. The left-hand plot also uses x = 0. relative to a displayed market.005 Limit price 0. and paid at the end of period T . an asset pays an uncertain dividend v that is +σ or −σ with equal probabilities. 16 y+ (p) y− (p) 15. Hendershott and Jones (2005) empirically study price discovery for exchange-traded funds (ETFs) when Island ECN stopped displaying its limit orders.1 −0. which operate alongside exchanges. 2. To simplify the analysis. t ∈ {1. that is.6 0. respectively.2 0.5 No informed participation 0. an informed trader prefers the exchange to the dark pool. As before. The left-hand plot shows the aggregate limit orders in the dark pool. Since Island ECN was the dominant market for affected ETFs. . Time is discrete.Figure 7: A dark pool as a nondisplayed limit order book. Under natural conditions. where y + (p) and y − (p) denotes the demand schedule and supply schedule.5 14 0. I drop endogenous information ac28 .01 −0. and realizations Z + = 31 and ¯ − Z = 30. The trading game ends immediately after the dividend payment. where T is deterministic. 3. hence improving price discovery.8 1 traders provide liquidity through limit orders when their demand schedules (i. This nondisplayed market. Model parameters are those of Figure 3. 5 Dynamic Trading This short section generalizes the basic intuition of Section 3 to a dynamic market. they conclude. The dividend is announced at the beginning of period T ≥ 2. }. informed traders avoid the dark pool. all equilibria have the property that.2 0. encourages informed traders to trade more aggressively on their information.5 Mass 0.8) = 0. σ = σ (0.e.4 x 0. The right-hand plot shows the range of σ for which the equilibrium of Proposition 4 exists.3 σ 0. after controlling for delay costs.01 0 0 15 14.8.005 0 0. relative to a liquidity trader. . In each period before the dividend payment. . limit orders) are hidden. a new set of informed traders and liquidity traders arrive. it differed from today’s equity dark pools.236.

Both venues are open in all periods. it costs zero to acquire information. To control for traders’ characteristics other than information. I now fix a cost type γ ≥ 0 and compare the venue choice of a type-γ informed 29 . Liquidity traders differ from each other in their delay costs. is deterministic. After execution of exchange orders in each period. is a special case of this double auction. Pt (v = −σ) (38) where Pt denotes the conditional probability based on Ft . simultaneously with the execution of exchange orders. The dark pool implements a double auction with a trade-at rule. Informed traders observe the dividend v and trade in the corresponding direction. which offers a price improvement of the exchange spread.) The mass of informed traders arriving in period t. then she incurs a delay cost of c(γ. respectively. the market maker announces the exchange buy volume and the exchange sell volume. a lit exchange and a dark pool operate in parallel. where c(γ. as in Section 4. seller) and a type-γ liquidity buyer (resp. Rt ) in period t if she fails to execute her order in that period. Midpoint crossing. before they execute their orders. In practice. but not including. the conditional distribution of asset value v at the beginning of period t is represented by the likelihood ratio Rt = Pt (v = +σ) . R0 = 1. I make the additional assumption that informed traders also incur positive delay costs. A trader only incurs delay costs after she arrives. The public information Ft at the beginning of period t consists of all exchange announcements prior to. The public does not observe v or the realizations of Z + (t) or Z − (t). a type-γ informed buyer (resp. A type-γ informed trader incurs the delay cost c(γ. If a liquidity trader of cost type γ does not trade in period t. The mass of liquidity buy orders and the mass of liquidity sell orders arriving in period t are Z + (t) > 0 and Z − (t) > 0. period t. with commonly known probability distributions. Rt ) in period t. (Equivalently. By construction. Any order sent to the exchange is immediately executed at the bid or the ask. as in Section 3. Thus. the exchange market maker posts a bid price Bt and an ask price At . At the beginning of period t. Rt ) is strictly increasing in γ for all Rt . Thus. seller) differ only in their information about v. µI (t) > 0. (39) V (Rt ) = σ(Pt (v = +σ) − Pt (v = −σ)) = Rt + 1 The dark pool executes orders in each period.quisition in this section. this cost may come from the opportunity cost of capital. The public’s conditional expectation of the asset value at the beginning of period t is therefore Rt − 1 σ. As before.

γ) − Wd (Rt . t. γ))] . The dark pool payoff difference Xd (Rt . t). It is in this “difference-in-difference” sense that the dark pool is more attractive to liquidity traders. Et (X(Rt+1 . Proposition 5. t. t. γ) − Wd (Rt . t. t. γ)]. t) − We (Rt . These payoffs do not depend on the cost type γ because exchange execution incurs no delays. γ) = 0. if Wd (Rt . respectively. Xe (Rt . See Appendix C. T . For any (Rt . t. Proposition 5 reveals that all equilibria must satisfy the restriction (44). before they make trading decisions. Wd (Rt . γ) ≥ Et [W (Rt+1 . t. the liquidity-versus-informed payoff difference Xd (Rt . and that the exchange is more attractive to informed traders. γ) = max [Xe (Rt . t)−We (Rt . t. Finally. γ) be the continuation values of the informed buyer and liquidity buyer. γ))] . for trading immediately on the exchange. controlling for delay costs. t. The following proposition characterizes equilibrium conditions under which. 30 (44) . γ) and Xd (Rt . γ) = X(RT . t + 1. reflects only the value of private information. For t = T . t) − We (Rt . t + 1. t) = V (Rt ) − At (40) (41) be the payoffs of a type-γ informed buyer and a type-γ liquidity buyer. t. the exchange payoff difference. (42) (43) where the three terms in the max( · ) operator represents a trader’s three choices: sending her order to the exchange. the Bellman Principal implies that W (Rt . at the beginning of period t. t). γ) and X(Rt . t. Because the exchange guarantees to execute all buy orders at the same price At . For t < T . I let We (Rt . γ). I let W (Rt . respectively. γ) = max [We (Rt . t) on the exchange. X(Rt . Xd (Rt . t). In any equilibrium. I let Wd (Rt . γ) ≥ Xe (Rt . t). t). sending her order to the dark pool. Proof. t). t) = σ − At Xe (Rt . t). t. reflects both the value of information and the execution risk. t + 1. The intuition for this result is as follows. t. γ) − Wd (Rt . provided that a informed buyer weakly prefers using the dark pool to delaying trade. and delaying trade. then Xd (Rt .buyer with that of a type-γ liquidity buyer. γ) in the dark pool is at least as high as the corresponding payoff difference Xe (Rt . by contrast. T . γ) be the corresponding continuation values of entering an order in the dark pool. Et (W (Rt+1 . W (RT .

31 . t. Prediction 4. I discuss empirical implications of the model and put them in the context of related empirical evidence. All else equal. more likely to pay a higher price. then dark pool market share can be decreasing in adverse selection (or volatility). t + 1. an informed buyer in the dark pool is less likely to fill her order and. If the level of adverse selection (or volatility) is high. 6. This execution risk is costly for the informed buyer in equilibrium as long as she prefers dark pool trading to delaying. The discussion follows two organizing questions. both in light of recent empirical evidence and in relation to the current policy debate on the impacts of dark pools on price discovery and liquidity. what are the relations between dark pool market share and observable market characteristics? Second. All else equal. as captured by Wd (Rt . conditional on a trade. dark pool market share is lower if the execution probability of dark pool orders is lower. First. All else equal. compared with dark pools that execute orders at the exchange midpoint. γ) ≥ Et [Wd (Rt+1 . γ)]. dark pool market share is lower for trading strategies relying on shorter-term information. I isolate this dark pool execution risk from the value of information by taking the “difference-in-difference” of payoffs in (44). All else equal. 6 Implications and Discussions This section discusses some implications of my results.Compared with a liquidity buyer. compared with strategies that trade individual stocks one at a time. if the level of adverse selection (or volatility) is low. informed participation in dark pools is higher if volatility is higher. what are the impacts of dark pool trading on price discovery and liquidity? For each question. then dark pool market share is increasing in adverse selection (or volatility). Prediction 2. Appendix B explicitly solves a dynamic equilibrium in a setting where traders arrive in Poisson times. Informed participation is higher in dark pools that allow more discretion in execution prices. The use of dark pools is also lower for trading strategies that trade multiple stocks simultaneously.1 Determinants of dark pool market share Prediction 1. Prediction 3.

To my knowledge. Related to this prediction. which is consistent with the notion that short-term strategies are best implemented in venues that guarantee execution. discourages both types of traders from partici¯ ¯ 22 pating in the dark pool. captured by a lower r or rx . He does not examine how non-execution probability relates to the market share of dark pools. Because dark pools cannot guarantee the This relation can be analytically proved for σ < σ in Proposition 1 and for σ < σ (x) in Proposition 4. depending on whether σ.finra. Using transaction data in two ¯ block-crossing dark pools (Liquidnet and Posit). 24 Recall from the model of Section 3 that if σ > σ .org/Industry/Regulation/Guidance/SECRule605/.Prediction 1 follows from the results of Section 3 and Section 4. then some informed traders participate in the dark pool. ¯ ¯ Related. and Werner (2011a) find that dark pool market share is negatively related to the order imbalance as a percentage of total volume and to the absolute depth imbalance on lit venues. Buti. Prediction 4 provides strategy-level implications on dark pool activity. Ye (2010) constructs a proxy for execution probability in eight dark pools from their SEC Rule 605 reports.g.24 and that volatilities and spreads are positively correlated with adverse selection. Using daily data collected by SIFMA from eleven anonymous dark pools in 2009. A lower execution probability. the value of private information. the latter half of Prediction 3 is not yet tested in the data. Prediction 3 suggests that dark pool orders are more informative on average when information asymmetry is severe. A trading strategy that follows the directions of dark pool orders is profitable when spreads are wide but not profitable when spreads are narrow. and Werner (2011a) and Ye (2010) find that dark pool market shares are lower when volatilities and spreads are higher. price impacts and effective spreads). To the extent that exchange spreads are proxy measures for adverse selection. Rindi. Prediction 1 is consistent with their findings. For more details of Rule 605 of Reg NMS. Strategies relying on shorter-term information have higher execution risks in dark pools because relevant information can become stale sooner. This prediction is consistent with recent evidence documented by Nimalendran and Ray (2012) in an anonymous dark pool. these findings are broadly consistent with (the latter half of) Prediction 2. Ready (2012) finds that institutions are less likely to route orders to dark pools when the level of adverse selection is higher. and studies the relationship between non-execution probability and market characteristics (e. Ready (2012) finds that the usage of block-crossing dark pools is lower for institutions with higher turnover. To the extent that at least some informed traders participate in dark pools in practice.23 Prediction 2 suggests that dark pool market share can be increasing or decreasing in the level of adverse selection (or volatility). see http://www. They infer the trading direction of each dark pool transaction by comparing the execution price with the prevailing market midpoint. Prediction 3 also suggests that orders in limit-order dark pools are more informative than those in midpoint dark pools. Prediction 3 is consistent with their results. ¯ 23 22 32 . Rindi. is below or above the threshold σ . Buti. In different samples.

and van Kervel (2011). higher price impacts. and smaller depths on lit markets. and Upson (2011) find that off-exchange (dark) order flows are less informative than exchange (lit) order flows. and Upson (2011). are associated with wider spreads. but not Buti. Weaver (2011) finds that higher levels of off-exchange trading in the U. Rindi. For these strategies.2 Effects of dark pools on price discovery and liquidity Prediction 5. and van Kervel (2011) find that higher market shares of dark trading—including dark pools and over-the-counter markets—are associated with higher price impacts. Jiang. however. All else equal. a higher dark pool market share increases the correlation of volumes across different stocks in lit exchanges. All else equal. Prediction 6. In Dutch equity markets. Prediction 7. a higher dark pool market share is associated with higher order informativeness. All else equal. Nimalendran and Ray (2012) document that following dark-pool transactions. Their results also indicate that exchange order flows become more informative as off-exchange order flows increase. Buti. Prediction 5 is consistent with empirical evidence from Degryse. Rindi. Proposition 3 provides sufficient conditions under which Prediction 5 holds. in that shares bought in dark pools tend to have low short-term returns and that shares sold in dark pools tend to have high short-term returns. Similarly. By contrast. and Werner (2011a) find that higher dark pool trading activity tends to be associated with lower spreads and lower return volatilities. and higher volatilities. This cost. partial execution in dark pools can be particularly costly.S. Using transaction data in an anonymous dark pool. wider spreads. higher quoted spreads.S. This cross-stock volume correlation is lower in dark pools than in lit exchanges. McInish. 6. Nimalendran and Ray (2012). McInish.simultaneous execution of trades in multiple stocks. higher realized spreads. bid-ask spreads tend to widen and price impacts tend to increase. we also expect dark pools to be less attractive for strategies tracking stock indices or “arbitraging” perceived mispricing among similar securities. Degryse. and Werner (2011a) or O’Hara and Ye (2011). which suggest a 33 . and higher price impacts of trades on the exchange. especially if the relative bid-ask spreads are already high. dark pool execution implies a positive adverse-selection cost. after adjusting for trading volumes in dark and lit markets. equity markets. Jiang. is lower than the exchange spread at the time of execution. de Jong. and Weaver (2011). de Jong. in U.

a dark pool may send an “indication of interest” (IOI). the addition of a dark pool concentrates informed traders on the exchange and improves price discovery. we should interpret these conflicting results with caution. which contains selected order information such as the ticker. and Werner (2011b) considers a setting where selected traders are informed of the state of the dark pool. Buti. 7 Concluding Remarks In recent years. side. which must be disseminated to the broad market immediately. This paper provides a simple model of dark pool trading and their effects on price discovery and liquidity. The Securities and Exchange Commission (2009) proposed to treat actionable IOIs—IOIs containing the symbol. Finkelshteyn. more “toxic”). In practice. this prediction is not yet tested in the data. One of these is information leakage. Rindi. O’Hara and Ye (2011) also conclude that higher fragmentation of trading is associated with faster execution. Prediction 6 can be viewed as the mirror image of Prediction 4. the volume correlation across stocks should be higher in lit venues than in dark pools. Prediction 7 on the adverse selection in dark pools is consistent with Sofianos and Xiang (2011). In this sense. Mittal (2008) and Saraiya and Mittal (2009) emphasize that short-term adverse selection in dark pools can reduce execution quality of institutional investors. Næs and Odegaard (2006) provide anecdotal evidence that filled orders in a dark pool are subject to shortterm losses. those strategies should have higher concentration in lit venues than in dark pools. to potential counterparties in order to facilitate a match. Johnson. Finally. and more efficient prices. lower transaction costs.better market quality. Since dark pools are less attractive to strategies that execute multiple stocks simultaneously. although they do not explicitly measure the costs of adverse selection. dark pools have become an important part of equity market structure. Conrad. 34 . To my knowledge. at the cost of reducing exchange liquidity. and Wahal (2003). For example. I show that under natural conditions. Given the wide variety of data samples used in these studies and the difficulty in completely correcting for endogeneity. and Domowitz. Besides price discovery and liquidity. size. and price of an order—as quotes. and Yegerman (2009) examine execution costs in dark pools. there are a few additional aspects of dark pools that contribute to their controversy. Consequently. who find that dark pools that have higher execution probabilities also have more severe adverse selection (that is. these dark pools are not completely dark. Brandes and Domowitz (2010).

it seems plausible that the lack of fair access can reduce trading opportunities and the welfare of excluded traders. results from Boni. Finally. 25 35 . 2010. Foster.25 For example. and Leach (2012) indicate that the exclusion of short-term traders in a dark pool (Liquidnet) improves the execution quality of institutional orders. Brown. Whether investors suffer from the lack of fair access can depend on perspective. whether customer orders are exposed to proprietary trading flows. In the United States. dark pools are opaque not only in their orders. what anti-gaming controls are in place. and Ramaswamy (2007) theoretically illustrate that setting a volume threshold in the dark pool—i. “Keynote Speech to the National Organization of Investment Professionals. how customers’ orders are routed.” Opaque operating mechanics of dark pools can make it more difficult for investors and regulators to evaluate the impact of dark pools on price discovery. and Litwin.” observes SEC Deputy Director James Brigagliano. liquidity. Gervais.” by James A.e. 2010). The International Organization of Securities Commissions (2010) also observes that “[l]ack of information about the operations of dark pools and dark orders may result in market participants making uninformed decisions regarding whether or how to trade within a dark pool or using a dark order. and market quality. a Greenwich Associates survey of 64 active institutional users of dark pools reveals that. On the one hand. on many occasions. Feng. how orders interact with each other. the dark pool executes orders only if trading interests on both sides of the market reach that threshold—can sometimes prevent impatient traders or informed traders from participating in the dark pool. and at what price orders are matched (Bennett. “some dark pools attempt to protect institutional trading interest by raising access barrier to the sell-side or certain hedge funds. but also in their trading mechanisms. April 19. dark pools are not required to provide fair access unless the dark pool concerned reaches a 5% volume threshold. For example. Brigagliano. Colon. On the other hand.Another consideration is fair access. dark pools do not disclose sufficient information regarding the types of orders that are accepted.

the market share of block-cross dark pools has declined from nearly 20% in 2008 to just above 10% in 2011. For this reason. and their respective market shares are plotted in Figure 8.4 0. another classification is provided by Tabb Group (2011). from around 50% to around 70%. 36 . and liquidity-provider platforms.S. The market share of liquidity-providing dark pools increased to about 40% around 2009. but then declined to about 20% in mid 2011. They categorize dark pools into block-cross platforms. Besides the three-way classification of dark pools discussed in Section 2. however.8 0. Continuous-cross dark pools have gained market share during the same period. do not cover the entire universe of dark pools. and the components of each category can vary over time.2 0 Block cross Continuous cross Liquidity providers 2009 2010 2011 Dark pools are also commonly classified by their crossing frequencies and by how they search for matching counterparties. As we can see. dark pools as fractions of total U. dark pool volume. The three types are summarized in the top panel of Table 2. estimated by Tabb Group.S. as illustrated in the bottom panel of Table 2. 1 0. The main features of these three groups are summarized in the top panel of Table 2. continuous-cross platforms. Tabb Group’s data. these statistics are noisy and should be interpreted with caution.Appendix A Institutional Features of Dark Pools This appendix discusses additional institutional features of dark pools and nondisplayed liquidity that are not covered in Section 2.6 0. Figure 8: Market shares of three types of U.

Liquidnet Electronic messages sent to potential matched counterparties Liquidnet Credit Suisse Crossfinder. LeveL. The bottom panel shows a classification by the crossing frequencies and the methods of finding counterparties. Goldman Sachs Owned by broker-dealers.Table 2: Dark pools classification by trading mechanisms. Direct Edge MidPoint Match Pipeline. Types Block cross Continuous cross Liquidity provider Classification by Tabb Group Examples Notes Liquidnet. with some randomization ITG POSIT Now. Instinet CBX. Barclays LX. Goldman Sachs Similar to the second group of the top panel. Getco limit order books or electronic market makers . Instinet Cross Similar to the first group of the top panel Credit Suisse Crossfinder. run as nondisplayed Sigma X. BIDS. Deutsche Bank SuperX Getco and Knight Same as the third group of the top panel 37 Types Scheduled Matching Continuous Advertized Negotiated Internal Classification by trading frequency and counterparty search Examples Typical features ITG POSIT Match. The top panel shows the classification of Tabb Group. Instinet US Crossing Cross at fixed clock times. POSIT Alert. Knight Link. Morgan Stanley MS LeveL is owned by a consortium of broker-dealers Pool. Sigma X.

27 Finally. Today. This sharp contrast in order sizes can be attributed to the use of algorithms that split “parent” orders into smaller “children” orders. For example.000 shares per order) and much larger than those of other broker dark pools (about 300 shares per order). as observed by the Securities and Exchange Commission (2010).28 The other source of 26 The Securities and Exchange Commission (2009) has recently proposed a rule requiring that alternative trading systems (ATS).26 Second. Ye (2010) finds that only eight U. in order to facilitate a match. The majority of dark pools cannot. all under the exchange’s MPID. dark pools can be uniquely identified by MPIDs from their Rule 605 reports to the SEC. For example. ECNs. it is generally not possible to assign a TRF trade to a specific off-exchange venue that executes the trade. Nonetheless. 28 There are exceptions. First. which aggregate trades executed by all offexchange venues—including dark pools. According to Rosenblatt Securities (2011). two block-size dark pools (Liquidnet and Pipeline) have an order size of around 50.” or TRFs. by which a broker-dealer handles customer orders as a principal or an agent (Securities and Exchange Commission. a broker-dealer may report customer-to-customer trades in its dark pool together with the broker’s own over-the-counter trades with institutions. there are two sources of nondisplayed liquidity that are usually not referred to as dark pools. Characteristics that distinguish dark pools also include ownership structure and order size. trades in an exchange-owned dark pool can be reported together with trades conducted on the exchange’s open limit order book. Order sizes can also vary substantially across dark pools. There are at least two reasons why high-quality data are lacking on dark pool trading in the United States. including dark pools. provide real-time disclosure of their identities on their trade reports. they are highly automated systems and rely 38 . whereas the latter typically involves broker-dealers as intermediaries.S. Thus. and broker-dealer internalization—into a single category. One is broker-dealer internalization. whereas a small fraction is owned by consortiums of broker-dealers or exchanges.000 shares. 2010).Aside from mechanisms such as midpoint-matching and limit order books. 27 For example. dark pool trades are reported to “trade reporting facilities. Because different trading mechanisms share the same MPID. all under the same MPID. dark pools often do not have their own identification numbers (MPID) for trade reporting. A crude way of distinguishing dark pools from broker-dealer internalization is that the former are often marketplaces that allow direct customer-to-customer trades. advertisement is sometimes used to send selected information about orders resting in the dark pool to potential counterparties. Similarly. which is larger than that of Posit (around 6. in the United States. knowing the MPID that executes a trade is insufficient to determine whether that trade occurred in a dark pool. dark pools acting like electronic market makers (like Getco and Knight) also provide liquidity by trading on their own accounts. most dark pools are owned by broker-dealers (with or without proprietary order flows).

becomes displayed. As before. t ≥ 0. for λF > 0. By competitive pricing. the bid price at any time t is the conditional expected asset value given the arrival of a new sell order at time t and given all public information up to. which are limit orders that are partially or fully hidden from the public view. say 1. which give rise to tractable stationary equilibria. 30 See http://www. A competitive and risk-neutral market maker on the exchange continually posts bid and ask prices for one unit of the asset. Examples include reserve (“iceberg”) orders and pegged orders. Different from Section 5.”) Each trader can buy or sell one unit of the asset. until the entire order is executed or canceled. Typically. information and traders in this section arrive at Poisson times. Once the displayed part is executed. and the market opens at time 0. a lit exchange and a dark pool operate in parallel. As before.nasdaqtrader. Any order sent to the exchange is immediately executed at the bid or the ask. an asset pays an uncertain dividend v that is +σ or −σ with equal probabilities.30 In particular. Liquidity traders. As in Section 5. Once an less on human intervention than. Upon arrival. and trade information is immediately disseminated to everyone. Nasdaq reports that more than 15% of its order flow is nondisplayed. time t. dealers arranging trades over the telephone. I do not consider endogenous information acquisition here. The time of the dividend payment is exponentially distributed with mean 1/λF . The ask price is set likewise. as in Glosten and Milgrom (1985). but before. pegged orders and hidden portions of reserve orders have lower execution priority than displayed orders with the same limit price.29 For example.aspx?id=DarkLiquidity 39 . Pegged orders are often fully hidden. arrive with an unwanted position in the asset whose size is either +1 or −1. 29 A reserve order consists of a displayed part. say. and a hidden part. Two types of risk-neutral traders—liquidity traders and informed traders—have independent Poisson arrivals with respective mean arrival rates of λL and λI . the same amount.nondisplayed liquidity is the use of hidden orders on exchanges. B Dynamic Trading with Stochastic Crossing This appendix explicitly characterizes a family of dynamic equilibria in which informed traders do not participate in the dark pool. Time is continuous. (Traders are thus “discrete. midpoint-pegged orders on exchanges are similar to dark pool orders waiting to be matched at the midpoint. equally likely and independent of all else.com/Trader. an informed trader observes v perfectly.800 shares. taken from the hidden part. The market maker also maintains a public “midpoint” price that is the conditional expected asset value given all public information up to but before time t. say 200 shares. who are not informed regarding the dividend.

the information content of a buy order is equal to that of a sell order. Pt (v = −σ | Q = 1) Pt (Q = 1 | v = −σ) Pt− (v = −σ) (49) In the equilibria characterized in this section.” that is. 40 . and midpoint prices.5µt (47) is the probability that an exchange trader arriving at t is “correct. I let λt be the time-t arrival intensity (conditional mean arrival rate) of traders of any type to the exchange.5 = 0. The dark pool accepts orders continually. the market maker immediately updates her bid. as in Section 3. so there is no need to specify them separately. The conditional expected asset value is. and let µt be the time-t conditional probability that an arriving exchange trader is informed. The likelihood ratio zt = qt 1 − qt (48) then represents the informativeness of a time-t exchange order. As in Section 5.31 For example. if a buy order hits the market maker’s bid at time t. buying if v = +σ and selling if v = −σ. R0 = 1. and an order sent to the dark pool is observable only by the order submitter. as in Section 5. ask. By construction. For analytical tractability. I assume that unmatched orders in the dark pool are immediately sent to the exchange market maker. The dark pool executes orders at the midpoint price and at the event times of a Poisson process with intensity λC that is independent of all else. Pt (v = −σ) (45) where Pt denotes the market maker’s conditional probability. Rt + 1 (46) To calculate the bid and ask prices. Then. the conditional likelihood ratio of v at time t is Rt = Pt (v = +σ) . V (Rt ) = σ(Pt (v = +σ) − P(v = −σ)) = Rt − 1 σ.5 + 0. then Bayes’ Rule implies that Rt = 31 Pt (v = +σ | Q = 1) Pt (Q = 1 | v = +σ) Pt− (v = +σ) = · = Rt− zt . who then executes these orders at the conditional expected asset value given all past public information and given the quantity and direction of unmatched orders from the dark pool. qt = µt + (1 − µt )0. Allocation in the dark pool is pro-rata on the heavier side.exchange order is executed.

Rt + 1 (Rt + 1)2 (54) By independent splitting of Poisson processes. J (51) (52) κj = 1. With probability κj and independently of all else. and where Rt− ≡ lims↑t Rs . (Rt + 1)2 (53) A liquidity seller of type j has the same flow cost cj because t γj Et [max(0. Without loss of generality. 41 .” t Because I look for stationary equilibria. the arrival intensities of type-j liquidity buyers and type-j liquidity sellers are both 0. Similarly. we focus on the strategies of informed buyers and liquidity buyers. the payoffs W (Rt ) and X(Rt ) depend on the public information Rt but not on time t. V (Rt ) is generally not identical to the −1 bid-ask midpoint. then Rt = Pt (v = +σ | Q = −1) Rt− = . whose payoffs are denoted W (Rt ) and X(Rt ). where the superscript “i” stands for “informed. a liquidity buyer of type j incurs a flow cost of cj = γj Et [max(0. Nonetheless. j=1 Before executing her order. V (Rt ) − v)] = γj 1 · Rt + 1 Rt − 1 2Rt − (−1) σ = γj σ. v − V (Rt ))] = γj t Rt − 1 Rt · 1− Rt + 1 Rt + 1 σ = γj 2Rt σ. Pt (v = −σ | Q = −1) zt (50) −1 To break even. an arriving liquidity trader incurs a cost of γj per unit of time for every unit of collateral support in her risky position. respectively. Because V ( · ) is nonlinear. for simplicity I refer to V (Rt ) as the “midpoint” price. For simplicity.5κj λL . (V (Rt zt ) + V (Rt zt ))/2. I use Ei [ · ] as a shorthand for Et [ · | v = σ].where Pt− denotes the market maker’s probability conditional on all exchange transactions up to but before time t. the market maker quotes a time-t bid price of V (Rt zt ) and a time-t ask price of V (Rt zt ). Liquidity traders must hold collateral equal to the expected loss on their unwanted risky positions. if an exchange sell order arrives at time t. where (κj )J and (γj )J are commonly-known constants and satisfy j=1 j=1 0 ≤ γ1 < γ2 < · · · < γJ−1 < γJ .

immediately after a dark pool crossing. that is. . enter orders in the dark pool. J}. Moreover. (58) γj < (λC + λF ) ze − 1 . . informed traders are relatively impatient. the market maker quotes a bid of V (Rt ze ) and an ask of V (Rt ze ). whereas liquidity traders are relatively patient. Informed traders trade on the exchange immediately upon arrival. as formally stated in the following lemma. At time t. 2. . Let Q be the direction of the next exchange order that arrives before the dividend payment. −1 4. underlie the partial separation between informed traders and liquidity traders in the equilibria of Proposition 6. 1. Lemma 1. Under the strategies stated in Proposition 6: 42 . A key step in the equilibrium solution of Proposition 6 is that informed traders expect the exchange price to move against them over time. Type-j liquidity traders.Proposition 6. . If the dark pool has not crossed by the time that the dividend is paid. Immediately after the dividend v is paid. 1 ≤ j < M . Type-j liquidity traders. 1 ≤ j < M. the market maker executes all outstanding orders at the cum-dividend price of v. For fixed integer M ∈ {0. Q = 1 denotes a buy order and Q = −1 denotes a sell order.5 0. trade immediately on the exchange upon arrival. define ze = Under the conditions λC < J i=M λI + 0. (55) κj λL 2λI λF . Thus. ze γj > (λC + λF )(ze − 1) + λI + i=M κi λL there exists an equilibrium in which: 1. J (56) (57) (ze − 1)3 . See Appendix C. the market maker executes all outstanding orders at a price of V (Rt ). Proof. 3. they cancel their dark pool orders and trade immediately on the exchange. but liquidity traders expect the exchange spread to narrow over time. These different expectations of future prices.5 J i=M J i=M κi λL κi λL . 2. M ≤ j ≤ J. √ ze ( ze + 1)2 M ≤ j ≤ J.

the more informative are exchange orders. without affecting the incentives of informed traders. in that Q V (Rt ) = Et [V (Rt ze )]. t (60) • The exchange ask price is a supermartingale for liquidity buyers. (59) • The exchange ask price is a submartingale for informed buyers. a higher arrival rate λF of information relaxes (56). informed traders avoid in the dark pool. making the dark pool more attractive to liquidity traders. based on the tightness of the incentive constraints (56)-(58). the more liquidity traders trade in the dark pool. Proposition 6 says that an informed trader trades immediately on the exchange if the crossing frequency λC of the dark pool is sufficiently low relative to the risk that her private information becomes stale. We now briefly discuss the comparative statics of the equilibria. Moreover. 2 (Rt + 1)2 (Rt ze + 1)(Rt ze + 1) (61) Proof. A liquidity trader sends her order to the dark pool if and only if her delay cost γ is sufficiently low compared to the potential price improvement obtained by trading at the market midpoint. On the other hand. This property is a dynamic analogue of the two-period equilibrium of Section 3. As long as λC is sufficiently low. suggesting that informed traders are less likely to trade in the dark pool if they face a higher risk of losing their information advantage. ze reflects the degree of information asymmetry on the exchange because it is the ratio of the mean arrival rate of traders in the “correct” direction versus the mean arrival rate of traders in the “wrong” direction. a higher λC relaxes (57) but tightens (58). because the exchange order informativeness ze is increasing in M . a higher delay cost γ makes the dark pool less attractive to liquidity traders. suggesting that informed traders are more likely to participate in the dark pool if the crossing frequency is higher. By contrast. in that Q V (Rt ze ) < Ei [V (Rt ze ze )]. a higher crossing frequency λC tightens (56). a higher λF makes the dark pool more attractive to liquidity traders by shortening their expected waiting time. as in (57)-(58). Third. First. Second. See Appendix C. In Proposition 6.• The asset value is a martingale for liquidity traders and the public. 43 . in that Q Et [V (Rt ze ze )] = V (Rt ze ) − 2 2Rt (ze − 1)3 σ.

We have ˜ ˜ − Xd (0) = 0 < −Xe (0). What remains to be shown is that the incentive-compatibility conditions (18)-(20) have a solution.1 Proofs Proof of Proposition 1 I define µI : [0. We first look for an equilibrium in which β = 0. and that the ¯ ¯ condition (64) is equivalent to σ > σ . β < 1. µI is uniquely determined by (17). µI (s) + (1 − G(1))µz ˆ (62) Given the value σ of information. For simplicity. So there exists a solution ∗ αd ∈ (0. ˜ − Xd (G(1)) = 1 − r ≥ ¯ µI (σ) ˆ ˜ = −Xe (G(1)).C C. µI (σ) + (1 − G(1))µz ˆ (63) and that a Case 2 equilibrium exists if and only if r >1− ¯ µI (σ) ˆ . For each αd . µI (σ) is the unique “knife-edge” mass of informed ˆ traders with the property that all informed traders and a fraction 1 − G(1) of liquidity traders send orders to the exchange. α0 = 0 and αe = 1 − αd . 44 . µI (σ) + (1 − G(1))µz ˆ where the second inequality follows from (63). ∞) → [0. Thus. For ˜ notational simplicity. By (15). µ] by ˆ ¯ µI (s) = µF ˆ ¯ (1 − G(1))µz s . otherwise. (17). I show that a Case 1 equilibrium exists if r ≤1− ¯ µI (σ) ˆ . that is. ¯ Clearly. The indifference condition of the marginal liquidity trader is given by (16). G(1)] that satisfies (16). µI (σ) + (1 − G(1))µz ˆ (64) Then I show that the condition (63) is equivalent to σ ≤ σ for some σ . the exchange spread would be zero and informed traders would deviate to trade on the exchange. To prove the proposition. we write the left-hand side of (16) as −Xd (αd ) and the right-hand ˜ side as −Xe (αd ). informed traders are indifferent between the exchange and the dark pool. and (62). ˜ ˜ we write the left-hand side of (18) as Wd (β) and the right-hand side of (18) as We (β). in equilibrium either β = 0 or 0 < β < 1. Now we look for an equilibrium in which β > 0.

and left-differentiability in Case 2. µI is unique determined by (20) and is increasing in β. the equilibrium in Case 1 exists if σ ≤ σ . Finally. Moreover. r+ and r− are implicitly defined by differentiable functions in each case of Proposition 1. for all αd and β > 0. ˜ ˜ ˜ ˜ We (β) > We (0) ≥ Wd (0) > Wd (β). ∞). suppose for contradiction that (64) does not hold. So there exists some r0 > 0 such that r− > r0 .For each β ≥ 0. Thus. 1) to (18). which implies that all informed traders wish to deviate to the exchange. there exists some unique volatility threshold σ at ¯ which (63) holds with an equality. αe . So there exists a solution ˜ β ∗ ∈ (0. Under condition (64) and for each αd > 0. differentiability refers ¯ σ ¯ to right-differentiability in Case 1 of Proposition 1. ˜ Wd (0) = r > 1 − ¯ where the first inequality follows from (63). ∗ So there exists a solution αd ∈ (0. they are continuous and differentiable in σ in each of the two intervals [0. by (62).2 Proof of Proposition 2 Because β. (17). G(1) − G r+ − r− r+ + r− > G(1) − G 1 − r0 1 + r0 > αd . That is. µ. contradicting β > 0. ¯ and the equilibrium in Case 2 exists if σ > σ . G(1)] to (19). increasing the value σ of information raises the knife-edge mass µ(σ) ˆ of informed traders. At the volatility threshold σ = σ . µI (σ) + (1 − G(1))µz ˆ ˜ ˜ Wd (1) = r− < 1 = We (1). The equilibria characterized by (18)(20) thus exist. (18) implies that in equilibrium r− is bounded away from 0. 45 . holding αd fixed. Because µI increases in β. So for sufficiently small αd > 0. αd . we see that Wd (β) < 0 ˜ and We (β) > 0. the solution β ∗ to (18) is unique for each αd . ¯ C. which in turn tightens the condition (63) under which informed traders avoid the dark pool. Then. as a function of αd . µI (σ) ˆ ˜ = We (0). To show that (64) is necessary for the existence of equilibria in which β > 0. S. Thus. and (62). σ ] and (¯ .

But (68) and (67) imply that d(βµI )/dσ = 0 = dµI /dσ at σ0 as well. Have a dark pool and σ > σ ¯ Now suppose that σ > σ . nor can it be zero. suppose σ otherwise. Total differentiation of (16)-(17) with respect to σ yields ¯ dG−1 (αd ) ∂(S/σ) dαd ∂(S/σ) dµI (1 − r) − ¯ − = 0. dαd /dσ cannot switch signs in [¯ . ∞). dµI /dσ > 0. and d(S/σ)/dσ > 0. β = 0. then (66) implies that dµI /dσ > 0 at σ0 . by (16). and dαd /dσ switches signs at some σ0 . We can show that dαd /dσ cannot switch signs in [¯ . ∞). To see why. Total differentiation of (18)-(20) with respect to σ yields r− ∂(1 − S/σ) 1 − αd ∂(βµI ) <0 d(βµI ) ∂(1 − S/σ) dµI βµI dαd = + r− 2 . Thus.Have a dark pool and σ ≤ σ ¯ For σ ≤ σ . Thus. dαd /dσ = 0. dαd /dσ > 0. at σ0 . some σ0 . I denote by r+ and r− the derivatives of r+ and r− with ¯ respect to βµI /αd . ∂µI dσ ∂αd dσ σ <0 >0 (66) where the first term of (65) is positive because of equilibrium selection. dαd ∂αd dσ ∂µI dσ >0 >0 (65) 1 − µF (σ − S) ¯ >0 ∂(σ − S) dαd S ∂ (σ − S) dµI = µF (σ − S) ¯ + µF (σ − S) 1 − ¯ . which contradicts (69). By continuity. σ 46 . ∂µI dσ ∂(βµI ) dσ σ >0 >0 (69) where the first term of (68) is positive because of equilibrium selection. If dαd /dσ ≤ 0 at. (r+ + r− )2 αd dσ >0 1−G r −r r+ + r− + − 2(r r − r r ) βµI dαd = −G 2 (r+ + r− )2 αd dσ >0 r+ − r− r+ + r− (68) 1 − µF (σ − S)σ ¯ >0 ∂(1 − S/σ) dµI ∂(1 − S/σ) d(βµI ) S = µF (σ − S)σ ¯ + µF (σ − S) 1 − ¯ . dσ ∂µI dσ αd dσ <0 <0 + − − + (67) 2(r+ r− − r− r+ ) 1 d(βµI ) . But then (65) cannot hold. We have r+ > 0 and r− < 0. say.

¯ d(βµI ) dσ = µI dβ dσ ≥ 0 =⇒ σ=¯ σ σ=¯ σ dαd dσ ≤ 0. (18) σ implies that d βµI d(S/σ) dr− =− = −r− > 0. By σ (68)-(69). then (71) implies that dµI /d¯ ≥ 0 at σ0 . But that r r contradicts (70).3 Proof of Proposition 3 Have a dark pool and σ ≤ σ ¯ For σ ≤ σ . βµI and µI are both strictly increasing in σ. dr+ d = r+ dσ dσ No dark pool The comparative statics for Corollary 1 are similar to that for the first case of Proposition 1 and are omitted. Then. C. S = σ · (S/σ). The total participation ¯ rate of liquidity traders in either the dark pool or the exchange is αd + αe = 1. raises αd and reduces αe = 1−αd . Moreover. Thus. ∞). adding a dark pool is equivalent to increasing r. for all σ ∈ [¯ . which is r r equivalent to an increase in r. βµI αd > 0. dr− d = r− dσ dσ βµI αd < 0. σ=¯ σ Because dαd /dσ cannot be zero. ∂µI d¯ r ∂αd d¯ r >0 <0 (71) where the first term on the left-hand side of (70) is positive because of the equilibrium selection. ∞). dαd /d¯ > 0 and dµI /d¯ < 0. Finally. by (68). higher than a market without a dark pool. Then. β = 0 and dβ/dσ ≥ 0.At σ = σ . ∂αd ∂αd d¯ r ∂µI d¯ r >0 >0 (70) ∂ dµI ∂(σ − S) dαd 1 − µF (σ − S) ¯ (σ − S) = µF (σ − S) ¯ . If dαd /d¯ ≤ 0 at any σ0 . Total differentiation of ¯ ¯ (16)-(17) with respect to r yields ¯ (1 − r) ¯ ∂G−1 (αd ) ∂(S/σ) dαd ∂(S/σ) dµI − = G−1 (αd ) + . 47 . by (17). it must be strictly negative for all σ ∈ [¯ . obviously increases in σ as well. dσ dσ dσ αd The spread itself. a lower µI implies a wider spread S on the exchange. Adding a dark pool.

Have a dark pool and σ > σ ¯ Now suppose that σ > σ . In a market with a dark pool, αe = 1 − G(1), a constant. ¯ Substituting it into (21) and we have µI < 1. µI + (1 − G(1))µz So the equilibrium αe without a dark pool resides in the interval (1 − G(1), 1). That is, adding a dark pool reduces αe . Moreover, adding a dark pool increases the exchange spread if and only if αe in the equilibrium of Corollary 1 is larger than (1 − G(1))/(1 − β), where β > 0 is determined in Proposition 1. By the equilibrium selection rule and by (18), αe > 1 − G(1) 1 − G(1) ⇐⇒ G−1 1 − 1−β 1−β > µI = 1 − r− , µI + µz (1 − G(1))/(1 − β) (72)

where the µI is given by µI = µF ¯ We rearrange (72) and obtain β< G(1) − G(1 − r− ) . 1 − G(1 − r− ) (1 − G(1))µz (1 − β)µI + (1 − G(1))µz .

On the other hand, because the left-hand side of (18) is decreasing in β and the righthand side is increasing in β, the above condition is equivalent to (23). As F (c) → 1 for all c > 0, (20) implies that µI → µ, a constant. Holding µI = µ ¯ ¯ − − fixed, we now show that if G (1 − r ) ≤ 0, then (23) holds for all r ∈ [0, r]. At ¯ r− = r, we have σ = σ and (23) holds by the definition of σ . At r− = 0, (23) also ¯ ¯ ¯ holds trivially. Take the first and second derivatives of the right-hand side of (23) with respect to r− and we obtain d[rhs(23)] µµz G (1 − r− ) ¯ = > 0, − dr [¯ + (1 − G(1 − r− ))µz ]2 µ d2 [rhs(23)] G (1 − r− )[¯ + (1 − G(1 − r− ))µz ] − 2µz [G (1 − r− )]2 µ = µµz ¯ < 0. − )2 d(r [¯ + (1 − G(1 − r− ))µz ]3 µ Thus, the right-hand side of (23) is concave and (23) holds for all r− ∈ [0, r]. ¯

48

C.4

Proof of Proposition 4

I prove this proposition in three steps. First, I calculate the execution price in the dark pool and the optimal limit prices chosen by liquidity traders. Second, I derive incentivecompatibility conditions under which informed traders choose not to participate in the dark pool. Finally, I prove the comparative statics and conditions under which the equilibrium exists. Step 1: Price p∗ of execution and optimal prices of limit orders I let y + : [−xS, xS] → [0, ∞) be the aggregate downward-sloping demand schedule of liquidity buyers in the dark pool, and let y − : [−xS, xS] → [0, ∞) be the aggregate upward-sloping supply schedule of liquidity sellers. For each p, y + (p) is the total mass of limit buy orders that have a limit price of p or higher, and y − (p) is the total mass of limit sell orders that have a limit price of p or lower. Because the dark pool crosses orders by price priority, its execution price p∗ is  xS,  if y + (p) > y − (p) for all p ∈ [−xS, xS].   p∗ = −xS, if y + (p) < y − (p) for all p ∈ [−xS, xS].    {p : y + (p) = y − (p)}, otherwise.

(73)

I proceed under the conjecture that the set {p : y + (p) = y − (p)} contains at most one element, in which case p∗ of (73) is uniquely well-defined. I later verify this conjecture. Once p∗ is determined, buy orders with limit prices above or equal to p∗ are matched, at the price of p∗ , with sell orders whose prices are at most p∗ . If there is a positive mass of buy or sell orders at the price p∗ , then traders setting the limit price p∗ are rationed pro-rata, as before. I now derive the optimal limit prices of liquidity traders in the dark pool, under the conjecture that the probability distribution of p∗ has no atom in (−xS, xS). This no-atom conjecture, verified later, implies that a liquidity trader quoting a price of p ∈ (−xS, xS) has her order filled with certainty (i.e. is not rationed) if p∗ = p. Thus, a liquidity buyer who has a delay cost of c ∈ [0, xS) and quotes a price of p in the dark pool has the expected payoff (negative cost) Xd (p; c) = −E I{p≥p∗ } p∗ + I{p<p∗ } c
p

= −c −
−xS

(p∗ − c) dH(p∗ ),

(74)

49

where I( · ) is the indicator function and H(p∗ ) is the cumulative distribution function of p∗ . Because there is no adverse selection in the dark pool, the execution cost for this liquidity buyer is either the payment p∗ or the delay cost c. Conjecturing that H(p∗ ) is differentiable with H (p∗ ) > 0 for p∗ ∈ (−xS, xS), properties that are also verified later, we obtain dXd (p; c) = −(p − c)H (p). (75) dp Because (75) shows that the sensitivity of expected payoff to the limit price p is positive for p < c and negative for p > c, the optimal limit price for the liquidity buyer is her delay cost c. Symmetrically, the optimal limit price for a liquidity seller with a delay cost of c ∈ [0, xS) is −c. This “truth-telling” strategy is also ex-post optimal, in that no one wishes to deviate even after observing the execution price. The first-order condition (75) also implies that xS is the highest limit price in the dark pool, and that −xS is the lowest limit price.32 Let y(p) be the downward-sloping demand schedule in the dark pool if Z + = 1. Because a limit price p ∈ [0, xS) is submitted by the liquidity buyer with the delay cost p, max(0, p) , −xS < p < xS. (76) y(p) = αd − G σ By symmetry, the liquidity buyers’ demand schedule and the liquidity sellers’ supply schedule in the dark pool are, respectively, y + (p) = Z + y(p), y − (p) = Z − y(−p). (77) (78)

Because the equation y + (p) = y − (p) has at most one root, we have verified our earlier conjecture that the dark pool execution price p∗ is uniquely well-defined. Given y(p), the execution price p∗ in the dark pool is  +xS,      +σG−1 αd 1 − −σG−1 α 1 −  d     −xS, if [αd − G
Z− Z+ Z+ Z− xS σ xS σ xS σ xS σ

]Z + ≥ αd Z − , Z + < αd Z − ≤ αd Z + , Z − < αd Z + ≤ αd Z − , ]Z − ≥ αd Z + . (79)

p =

, if αd − G , if αd − G if [αd − G

If the maximum limit price were lower, say p0 < xS, then a liquidity buyer with a delay cost of p0 + for some small > 0 would deviate to the dark pool and quote p0 + . This deviating buyer has an execution probability of 1 and pays at most p0 + < xS ≤ S, which is better than execution on the exchange. The argument for the lowest limit price is symmetric.

32

50

this liquidity trader has a delay cost of G−1 (αd )σ. and strictly increasing on (−xS. c) = −(1 − rx )(c − xS). Thus. (36) and (37) characterize an ∗ equilibrium. the dark pool execution price p∗ has a mean of zero. ¯ As before. So we must have Xd (xS. an equilibrium is determined by a marginal liquidity trader who is indifferent between the dark pool and the exchange. r ¯ (83) (82) That is. in order to get the highest priority. H(p∗ ) is continuous. an informed buyer behaves in the same way as does a liquidity buyer who has a delay cost of σ.Because the total trading interest Z + of liquidity buyer and the total trading interest Z − of liquidity sellers are identically distributed. We now look for conditions under which. ¯ (αd − G(xS/σ))Z + and an expected payoff. c) = σ¯x + c(1 − rx ) ≤ σ = We − Xe . this deviating informed ¯ trader sets the highest limit price xS. ¯ (81) . of Xd (xS. Wd − Xd (xS. given the delay cost c. αd ≤ G(1). xS). In 51 . an informed buyer who deviates to the dark pool also has the crossing probability rx . G−1 (αd )) = −S. A liquidity buyer quoting the limit price xS in the dark pool has an execution probability of αd Z − rx = E min 1. If informed traders do not participate in the dark pool. Because informed traders avoid the dark pool with probability 1 in the conjectured equilibrium. differentiable. for any delay cost c ≤ σ. Moreover. (80) This expected payoff calculation follows from the fact that E(p∗ ) = 0 and the fact that failing to cross in the dark pool incurs a delay cost of c but saves the payment xS. as conjectured earlier. Step 2: Incentive conditions for participation What remains to be shown are the incentive-compatibility conditions of liquidity traders who set the limit price xS or −xS in the dark pool. or (36). as well as the incentive-compatibility condition of informed traders. By the differentiability of G and of the distribution function of Z − /Z + . Her expected profit in the dark pool is thus Wd = σ − (1 − rx )(σ − xS). in equilibrium. Given αd . who avoid the dark pool.

I define K(x) implicitly by ¯ ¯ (1 − xK(x)) 1 − E min 1. 1]. if dαd /dσ ≤ 0 at some σ0 . given K(x).this equilibrium. Moreover. I define µ∗ (x) by I µ∗ (x) I ¯ = K(x). Step 3: Comparative statics and conditions for the existence of equilibria I now calculate the comparative statics. And given the equilibrium. ¯ On the other hand. Total differentiation of (36) and (37) with respect to σ yields ∂[lhs(36)] ∂[rhs(36)] − ∂αd ∂αd >0 dαd + dσ ∂[lhs(36)] ∂[rhs(36)] − ∂µI ∂µI <0 dµI = 0. dσ (84) 1− ∂[rhs(37)] ∂µI >0 dµI ∂[rhs(37)] dαd S = + µF (σ − S) 1 − ¯ dσ ∂αd dσ σ <0 >0 . G(1)Z − ¯ [G(1) − G(xK(x))]Z + ¯ = K(x). But this contradicts (85). Thus. β = 0. assuming the existence of an equilibrium. µ∗ (x) + (1 − G(1))µz I 52 . the comparative statics with respect to σ follow. and then show conditions under which the stated equilibrium exists. (86) ¯ This K(x) is uniquely well-defined because the left-hand side of (86) is decreasing in ¯ ¯ K(x) and the right-hand side is strictly increasing in K(x). then (84) implies that dµI /dσ ≤ 0 at σ0 as well. the dark pool execution price p∗ and the optimal limit prices follow from calculations done in Step 1 of the proof. (85) As before. total differentiation of (86) with respect to x yields ∂[lhs(86)] ∂[lhs(86)] ¯ − 1 K (x) + = 0. Now I characterize the condition for the existence of an equilibrium and the thresh¯ old volatility σ (x). ¯ ∂x ∂ K(x) <0 <0 ¯ So we have K (x) < 0. For x ∈ [0.

For αd = G(1). the left-hand side of (36) is negative. t. t. γ) the “auxiliary payoff” of a type-γ liquidity buyer who “imitates” the strategy of a type-γ informed buyer. µI + (1 − G(1))µz ¯ and. there exists a solution αd ∈ (0. by construction. Therefore. Because K ¯ ¯ What remains to be shown is that. γ) ≥ X(Rt . Let Xd be the dark pool ˆ− payoff of the imitating buyer conditional on v = +σ. (1 − xK) 1 − E min 1. That is. by the definition of K(x). whereas the right-hand side is strictly positive.5 Proof of Proposition 5 Suppose that Wd (Rt . ¯ + (1 − G(1))µz µ∗ (x) I ¯ Because µ∗ (x) is strictly increasing in K(x) and because σ (x) is strictly increasing in ¯ I ∗ ¯ ¯ (x) < 0. σ (x) < 0. αd =G(1) (1 − G(1))µz σ . Thus. t + 1.and define σ (x) by ¯ ¯ µ∗ (x) = µF I (1 − G(1))µz σ (x) . so X(Rt . the imitating buyer behaves as if v = +σ. t. That is. Clearly. I denote by ˆ X(Rt . G(1)) to (36). µI is uniquely determined by (37). ˆ Clearly. in the calculations below I suppress the cost type γ and likelihood ratio Rt as function arguments. t. For sufficiently small αd . γ) ≥ Et [W (Rt+1 . and an equilibrium exists. and γ. For notional simplicity. Suppose that the informed buyer enters an order in the dark pool at time t. at αd = G(1). µI (x). such imitation is suboptimal for the liquidity buyer. γ)] in an equilibrium. G(1)Z − [G(1) − G(xK)]Z + > K. σ (x) is strictly increasing in K(x). once αd is determined. an equilibrium characterized by ¯ Proposition 4 exists. I K≡ S µI = µI + (1 − G(1))µz σ ¯ ≤ K(x). the left-hand side of (36) is weakly higher than the right-hand ∗ side. The ˆ+ imitating liquidity buyer does the same. and let Xd be the dark pool 53 . C. (37) implies that µI = µF ¯ which is no larger than µ∗ (x). for σ ≤ σ (x). γ) for all Rt . t.

Informed buyers have either a zero or positive mass in the dark pool in period t. we have ˆ Xd (t) − Wd (t) = Rt ˆ + 1 ˆ Xd (t) + X − (t) − Wd (t) Rt + 1 Rt + 1 d 1 ˆ− ˆ+ = Xd (t) − Xd (t) . Xd (t) − Wd (t) ≥ Xd (t) − Wd (t) ≥ − Rt + 1 ˆ+ ˆ− Now I prove that Xd (t) − Xd (t) ≤ 2σ. I denote by C the event that the imitating buyer’s order is crossed in the dark pool. + ˆ+ ˆ ˆ Xd (t) − Et (X + (t + 1)) = kt [σ − Et (p∗ | C. Then. we have + + ˆ ˆ+ Xd (t) = kt [+σ − Et (p∗ | C. Rt + 1 where the last equality follows from the fact that. d t t ˆ ˆ where p∗ is the execution price in the dark pool in period t. Xd (t) ≥ Et (X + (t + 1)). v = −σ). conditional on v = +σ and v = −σ. conditional on the true dividend. ˆ ˆ X − (t) = k − [−σ − Et (p∗ | C. t) − We (Rt . − kt ≡ Pt [C | v = −σ]. v = +σ) − Et (X + (t + 1))] ˆ ≤ k − [σ − Et (p∗ | C. Et (p∗ | C. t). and X + and X − are the imitating buyer’s payoffs conditional on v = +σ and v = −σ. v = −σ)] + (1 − k − )Et (X − (t + 1)). I define Xe and Xe similarly. the ˆ+ expected payoff Wd of the informed buyer is the same as the payoff Xd of the imitating ˆ+ ˆ− liquidity buyer. so + − kt ≤ kt . t 54 . Thus. Then. respectively.ˆ+ ˆ− payoff of the imitating buyer conditional on v = −σ. v = +σ) ≥ Et (p∗ | C. respectively. then 2σ ˆ = Xe (Rt . be the crossing probabilities of the imitating buyer in the dark pool in period t. If we can show that Xd (t) − Xd (t) ≤ 2σ. ˆ+ ˆ Because Wd (t) ≥ Et [W (t + 1)]. v = +σ)] + (1 − kt )Et (X + (t + 1)). v = −σ) − Et (X + (t + 1))]. and let + kt ≡ Pt [C | v = +σ].

(87) ˆ+ ˆ− ˆ ˆ I now prove that Xd (t) − Xd (t) ≤ 2σ and that X + (t) − X − (t) ≤ 2σ by induction. − + ˆ For all t < T . Then. (87) implies + − that Xd (t) − Xd (t) ≤ 2σ. Xe (t) − Xe (t) = 2σ. That is. q is the probability that an arriving exchange order is in the same direction as informed orders. the probability that v = +σ is Rt /(Rt + 1). By (87). X + (T − 1) − X − (T − 1) + − − + = max Xe (T − 1) − Xe (T − 1). Under a liquidity trader’s belief. the probability that the next exchange order is a buy order is Rt 1 Rt ze + 1 q+ (1 − q) = (1 − q) . which completes the proof. Let q be implicitly defined by ze = q/(1 − q). Xd (T − 1) − Xd (T − 1) ≤ 2σ. C. Xd (T − 1) − Xd (T − 1). ET −1 (X + (T ) − X − (T )) ≤ 2σ. + − + − X + (t) − X − (t) = max Xe (t) − Xe (t). Because v is revealed in period T . Because the venue choice of the imitating liquidity buyer does not depend on realizations of v. Et (X + (t + 1) − X − (t + 1)) ≤ 2σ. Rt + 1 Rt + 1 Rt + 1 55 . suppose that X + (t + 1) − X − (t + 1) ≤ 2σ. ˆ ˆ For the induction step. the probability that the next exchange order is a sell order is −1 Rt 1 Rt ze + 1 (1 − q) + q=q .6 Proof of Lemma 1 Given the public information Rt . X + (T ) = ˆ ˆ+ ˆ− X − (T ) = 0. Thus.which implies that − − ˆ ˆ ˆ+ ˆ− Xd (t) − Xd (t) ≤ 2σkt + (1 − kt )Et [X + (t + 1) − X − (t + 1)]. Xd (t) − Xd (t). Rt + 1 Rt + 1 Rt + 1 Similarly.

the exchange ask price is a submartingale for informed buyers. direct calculation gives 2 2Rt (ze − 1)3 σ. The Hamilton-Jacobi-Bellman (HJB) equation of an i=M 56 . Note that the identity (88) holds for all Rt . Finally. the exchange ask price is a supermartingale and the exchange bid price is a submartingale. the expected midpoint price is a martingale for liquidity traders. including the informed trader’s likelihood ratio ∞. The quoting strategy of the market maker simply follows from risk neutrality and zero profit. Using (88) again. C. we have that V (Rt ze ) = σ Rt ze + 1 −1 Rt ze ze + 1 2 Rt ze ze + 1 2 = (1 − q) 1− σ+q 1− −1 Rt ze + 1 Rt ze ze + 1 Rt ze + 1 Rt ze ze + 1 2 2 −1 σ + qze σ < (1 − q)ze 1 − 1− −1 Rt ze ze + 1 Rt ze ze + 1 Q = Ei [V (Rt ze ze )]. t 1− 2 σ That is.7 Proof of Proposition 6 I prove Proposition 6 by direct verification. Under the proposed equilibrium strategy. for liquidity traders. 2 (Rt + 1)2 (Rt ze + 1)(Rt ze + 1) −1 2 2Rt (1 − ze )3 Q −1 −1 σ. that is.We can verify that identity −1 Rt ze + 1 Rt ze + 1 −1 V (Rt ze ) + q V (Rt ze ) Rt + 1 Rt + 1 −1 Rt ze + 1 2 Rt ze + 1 =(1 − q) 1− σ+q Rt + 1 Rt ze + 1 Rt + 1 2 = 1− σ. Rt + 1 (1 − q) 1− 2 −1 Rt ze +1 σ (88) Q which implies Et [V (Rt ze )] = V (Rt ). Et [V (Rt ze ze )] = V (Rt ze ) + −1 −2 (Rt + 1)2 (Rt ze + 1)(Rt ze + 1) Q Et [V (Rt ze ze )] = V (Rt ze ) − That is. the total arrival intensity of exchange orders is λt = λI + J κi λL .

informed buyer is W (Rt ) = max σ − V (Rt ze ). • σ − V (Rt ). the profit if the next event is a dark pool cross. Q λt Et [X(Rt ze )] − cj t . the profit if the next event is the dividend payment. for all t and all realizations of random variable Rt . it is sufficient to verify that. Now we turn to a type-j liquidity buyer. a sufficient condition t for (89) is σ − V (R) λC + λF > sup λC = ze λC . Q λt Ei [W (Rt ze )] + λC (σ − V (Rt )) t . that is. Q λt Et [C(Rt ze )] + cj t . as implied by 57 . σ − V (Rt ze ) > Q λt Ei [σ − V (Rt ze ze )] + λC (σ − V (Rt )) t . If 1 ≤ j < M . it suffices to verify Q Q λt Et [V (Rt ze ze ) − V (Rt ze )] + cj t V (Rt ze ) − V (Rt ) > . depending on j. the expected profit for informed buyers to trade on the exchange is a Q supermartingale. λt + λC + λF There are two cases. the expected profit if the next event is the arrival of an exchange t order. to verify that C(Rt ) < V (Rt ze ) − V (Rt ). σ − V (Rze ) R∈(0. • 0. λt + λC + λF (90) Q Q Q where C(Rt ze ) is replaced by the higher cost of V (Rt ze ze ) − V (Rt ze ). σ − V (Rt ze ) > Ei [σ − V (Rt ze ze )]. λt + λC + λF (89) By Lemma 1. whose HJB equation is X(Rt ) = max −(V (Rt ze ) − V (Rt )). To verify that W (Rt ) = σ − V (Rt ze ). λt + λC + λF where the profit of immediate trading on exchange is σ − V (Rt ze ) and the expected profit of trading in the dark pool is the weighted sum of: Q • Ei [W (Rt ze )]. Thus. λt + λC + λF Her cost of liquidation C(Rt ) satisfies the HJB equation C(Rt ) = −X(Rt ) = min V (Rt ze ) − V (Rt ).∞) which simplifies to (56).

58 . If M ≤ j ≤ J. Because the ask spread is a supermartingale for liquidity buyers (Lemma 1). 2 (R z + 1)(R z 2 + 1) (Rt + 1) (Rt + 1)2 t e t e (91) (λC + λF ) (R + 1)(ze − 1) Rze + 1 (ze − 1) . it suffices to verify Q Q λt Et [V (Rt ze ze ) − V (Rt ze )] + cj t .∞) λt R(ze − 1)3 2 (Rze + 1)(Rze + 1) = (λC + λF )(ze − 1) + λt The argument for sellers is symmetric and yields the same parameter conditions.∞) 2 2Rt (ze − 1)3 2Rt σ + γj σ.the conjectured equilibrium. by Lemma 1.∞) γj · 2R/(R + 1)2 · σ V (Rze ) − V (R) = γj ze . V (Rt ze ) − V (Rt ) < λt + λC + λF that is. (λC + λF )[V (Rt ze ) − V (Rt )] < −λt A sufficient condition for (91) is γj > sup R∈(0. √ ze ( ze + 1)2 3 + sup R∈(0. ze − 1 which simplifies to (57). to verify that C(Rt ) = V (Rt ze ) − V (Rt ).∞) cj t V (Rze ) − V (R) = sup R∈(0. a sufficient condition for (90) is λC + λF > sup R∈(0.

d. and γ has c. trade in the dark pool.r Dark pool crossing probabilities conditional on informed traders being on the same and opposite side. H Xd (p. and trades in the dark pool. Vs Period-1 realized buy volume and sell volume on the exchange. σ µ. on the exchange. and defer trading. defined by (62) We . maximum (minimum) dark pool price is xS (−xS) Aggregate demand schedule in the dark pool if Z + = 1 Aggregate dark pool demand and supply schedules. αd . respectively β Fraction of informed traders who trade in the dark pool S Exchange (effective) spread. φ 2 µz . respectively Dark pool transaction price p∗ has a c. Z −.) of information-acquisition cost Signed informed trading interests: Y = sign(v) · µI Liquidity buy quantity Z + and liquidity sell quantity Z − have p. α0 Variables Introduced in Section 4 Strictness of trade-at rule. bid is −S and ask is S r ¯ Dark pool crossing probability if no informed traders go to the dark pool − + r .f. σ z c. µI ¯ F Y Z +.f. Variable Description Variables Introduced in Section 3 Asset value v is either +σ or −σ. respectively σ ¯ Maximum volatility for which informed traders avoid the dark pool µI (σ) ˆ Knife-edge mass of informed traders. V Expected volumes in the dark pool.D List of Model Variables This appendix summarizes key variables used in Section 3 and Section 4. φ Total mean and variance of liquidity trading interests Z + + Z − Delay cost of a liquidity trader is c = σγ per unit of asset.d. c) rx ¯ σ (x) ¯ 59 . for σ > 0 Total masses of for-profit traders and informed traders Cumulative distribution function (c. γ.f.d. G Fractions of liquidity traders who trade on the exchange. y− p∗ . trades on the exchange. Xe . Xd (c) Per-unit payoff of a liquidity buyer with a delay cost of c who defers trading. G αe . and both. respectively R1 Period-1 log likelihood ratio of {v = +σ} versus {v = −σ} P1 Period-1 closing price on the exchange I(β.d. of H Dark pool payoff of a liquidity buyer with the limit price p and per-unit delay cost c Dark pool crossing probability of a liquidity buyer with the limit price xS Maximum volatility for which informed traders avoid the dark pool x y y+. Ve . respectively Vd . Wd Expected profits of an informed buyer on the exchange and in the dark pool X0 (c). respectively v.f. αe ) Signal-to-noise ratio of period-1 exchange order flow Vb .

B. 2010.. and Pedersen. I. Journal of Finance 62: 2235–2274. and Pfleiderer.. C. Feng. L. and George. Wen. J. and Goldstein. Market Microstructure: The Impact of Fragmentation under the Markets in Financial Instruments Directive. and Viswanathan. 2007. B. S. Colon. Y. CFA Institute 2009. Review of Financial Studies 1: 3–40. Bennett.. Buti. Illuminating the New Dark Influence on Trading and U. Rindi. T. A. Brandes. 2012. J. Dark Pool Exclusivity Matters.References Admati. J. Journal of Trading 2: 108–113. D. Working paper Fisher College of Business. Annual Reviews of Financial Economics. Brown. Buti. and Leach.. and Domowitz. Working paper University of Houston... Buti. Working paper. Journal of Finance 60: 1825–1863. P. Working paper. K. J. Boulatov. G. Dynamic Dark Pool Trading Strategies in Limit Order Markets. S. Rindi. Carlin. Market Structure. H. Carrie. S. 2011. I. Technical report CFA Institute.. B. Forthcoming. Edmans.. Securities Trading when Liquidity Providers are Informed. Predatory Trading. R.. Ohio State University. M. L... 2011a. I. 2011b. S. Lobo. C. M.S. I. Boni. M. A.. Alternative Trading Systems in Europe: Trading Performance by European Venues Post-MiFID. Working paper. P.. 2010.. C. I. and Werner. M.. J. A Theory of Intraday Patterns: Volume and Price Variability. and Werner. Episodic Liquidity Crisis: Cooperative and Predatory Trading. Tick Size Regulation.. B. Intermarket Competition and Sub-Penny Trading. Y. J. 2008. and Litwin. Rindi. 2012. 2010. 1988.. Journal of Trading 5: 10–13.. I. and Werner. Diving into Dark Pools. Butler. The Real Effects of Financial Markets.. Journal of Trading 3: 40–55. Brunnermeier. Bond. 2007. Ten Questions Every Institution Should Ask Their Dark Pool Providers. Liquidity Aggregation: What Institutional Investors Need to Know. S. A. 60 .. Journal of Trading 5: 17–30. 2005. M.

Journal of Finance 51: 811–833. and Yegerman.. Working paper. Journal of Financial Economics 70: 99–134. M. and van Kervel. 1996. Review of the Markets in Financial Instruments Directive (MiFID).. The Benefits of Volume-Conditional Order-Crossing. M. S. Glosten.. An Optical Tour of Dark Pool Trading Performance.. and Regulation. Multimarket Trading and Market Liquidity. Consulation paper 23-404 Canadian Securities Administration and Investment Industry Regulatory Organization of Canada.. Ask and Transaction Prices in a Specialist Market with Heterogenously Informed Traders. V. Hendershott. R. Institutional trading and alternative trading systems. D. and Wuyts. Recent Trends in Trading Activity and Market Quality.. S. H. I. and Other Development in Market Structure in Canada. Public consultation European Commission. H. Dark Pools. Review of Financial Studies 18: 743–793. V.. A.. Review of Financial Studies 4: 483–511. Hendershott. Journal of Financial Economics 14: 71–100.... Fragmentation. CSA/IIROC 2009.Chordia. The Impact of Dark and Visible Fragmentation on Market Quality. L. R. M. 61 . Gervais. F. and Subrahmanyam. 2007. K. and Milgrom. N.. T. and Mendelson.. G. and Ramaswamy. Journal of Financial Economics 91: 319–338. Foster. Roll. B. T. 1991. Dynamic Order Submission Strategies with Competition between a Dealer Market and a Crossing Network.. J. Easley. Cul de Sacs and Highways. and Wahal. Conrad. Island Goes Dark: Transparency. de Jong. Domowitz.. C. Chowdhry. Bid. Journal of Finance 55: 2071–2115. 2009. Dark Orders. Keifer. Johnson. and O’Hara. 1985. 2000. Cream-Skimming or Profit-Sharing? The Curious Role of Purchased Order Flow. Degryse. H. Working paper. T. H. I.. 2009. D. European Commission 2010.. and Jones. and Nanda. M. K. M. Degryse. 2005. Crossing Networks and Dealer Marekts: Competition and Performance.. 2011. 2003. Journal of Trading 4: 16–22. Journal of Financial Economics 101: 243–263. 2011. Finkelshteyn. R. P. Van Achter.

Trading Volume and Asset Liquidity. Parlour. Chapter 3.. T. Continuous Auctions and Insider Trading. M. J. H. C. 2011. 2010. and Ray. O’Hara. Final report Technical Committee of the International Organization of Securities Commissions... C. B. Quarterly Journal of Economics 104: 255–274. Mittal. Journal of Financial Markets 9: 79–99. Econometrica 53: 1315– 1335.International Organization of Securities Commissions 2010. Nimalendran. Working paper University of Florida. S. P. 2006. A. D. Principles for Dark Liquidity. International Organization of Securities Commissions 2011. 62 . Handbook of Financial Intermediation and Banking. 4Myeloma Press. 2010.. 2011. Consultation report Technical Committee of the International Organization of Securities Commissions. Are You Playing in a Toxic Dark Pool? A Guide to Preventing Information Leakage. and Odegaard. A Dominant Strategy Double Auction. Why Fragmented Markets Have Better Market Quality: The Flight of Liquidity Order Flows to Off Exchange Venues. R. 2008. 2012. Journal of Financial Economics 100: 459–474. A. Joint CFTC-SEC Advisory Committee 2011. Limit Order Markets: A Survey. Working paper. Algorithmic Trading & DMA: An introduction to direct access trading strategies: . Næs. M. Johnson. M. M.. Informational Linkages Between Dark and Lit Trading Venues. 1992. Jiang. 1985. London. A. Issues Raised by Dark Liquidity. R. 2008. Is Market Fragmentation Harming Market Quality?. Pagano. Recommendations regarding Regulatory Responses to the Market Event of May 6. MacAfee. McInish. Summary report. B. Kyle. S.. and Upson. Equity trading by institutional investors: To cross or not to cross?. and Seppi. and Ye. 1989. The Journal of Trading 3: 20–33. J. Journal of Economic Theory 56: 434–450.

2010. Off-Exchange Reporting and Market Quality in a Fragmented Market Structure. M.Ready. Let there be light.. Technical report. 2012. Working paper Investment Technology Group. Part Two. The exact law of large numbers via Fubini extension and characterization of insurable risks. Securities and Exchange Commission 2010. 2009. and Xiang. Federal Register 74: 61207–61238. Transaction Costs and Market Share of the Crossing Network. Securities and Exchange Commission 2005. Regulation NMS. Tabb Group 2011. Federal Register 75: 3593–3614. Sofianos. 2011.. M. Sun. Market structure analysis. Proposed Rule. Working paper University of Illinois. Rosenblatt Securities 2011. H. Journal of Economic Theory 126: 31–69. Federal Register 70: 37496–37644. Y. Saraiya. 2011. J. 2006. Goldman Sachs. J. A Glimpse into the Dark: Price Formation. Working paper University of Illinois. Working paper. D. Ye. Non-execution and Market Share of Crossing Networks. Ye. Dark Pool Races. N. and Mittal. Concept Release on Equity Market Structure. Understanding and Avoiding Adverse Selection in Dark Pools. Working paper University of Wisconsin-Madison. 2011. 63 . Working paper. G. Proposed Rule. Weaver. Determinants of Volume in Dark Pools. M. Regulation of Non-Public Trading Interests. Securities and Exchange Commission 2009. Liquidity Matrix.