International Review of Financial Analysis xxx (xxxx) xxxx

Contents lists available at ScienceDirect

International Review of Financial Analysis

journal homepage: www.elsevier.com/locate/irfa

Brokered versus dealer markets: Impact of proprietary trading with

transaction fees
Katsumasa Nishidea, Yuan Tianb,*
a
Graduate School of Economics, Hitotsubashi University, 2-1 Naka, Kunitachi-shi, Tokyo 186-8601, Japan
b
Faculty of Economics, Ryukoku University, 67 Tsukamoto-cho, Fukakusa, Fushimi-ku, Kyoto 612-8577, Japan

ARTICLE INFO ABSTRACT

Keywords: In this study, we consider a one-period financial market with a dealer/broker and an infinite number of in-
Proprietary trading vestors. While the dealer who trades on his own account (with proprietary trading) simultaneously sets both the
Dealer/brokered market transaction fee and the asset price, the broker who brings investors' orders to the market (with no proprietary
Transaction fees trading) sets only the transaction fee, given that the price is determined according to the market-clearing
condition among investors. We analyze the impact of proprietary trading on the asset price, transaction fee,
JEL classification:
trading volume, and the welfare of investors. We find that the bid and ask prices set by the dealer who can
D53
engage in proprietary trading are more favorable to average investors. As a result, both the trading volume and
G12
D42 the transaction fee increase, and social welfare improves.

1. Introduction crisis, financial authorities in several countries have either adopted or

are considering adopting regulatory measures on investment banking,
There was an explosive growth of proprietary trading, in which including the Volcker rule in the US, the Liikanen Report to the EU, and
banks gambled with large chunks of their own capital and earned large the proposals of the Vickers Commission for the UK.1
profits before the 2007–2009 crisis. For example, it is reported that in As seen in Table 1, which concerns proprietary trading, while there
2006, Lehman Brothers generated 58% of its revenue from proprietary is a consensus on prohibiting principal trading (trading for banks' own
transactions, up from 33% in 1998. However, nearly every large fi- profit rather than on behalf of customers, including speculation and
nancial institution that stumbled during the financial crisis had billions manipulation), the proposals differ in whether or not to allow pro-
of dollars in proprietary-trading or hedge-fund losses. Lehman Brothers prietary trading as market-making activities (trading on behalf of cus-
lost more than $32 billion from proprietary trading and principal tomers by using banks' own accounts).2
transactions during 2008 and a half leading up to the financial crisis. The question of whether to ban proprietary trading as a market-
Morgan Stanley had a nearly $4 billion loss in proprietary trading in the making activity is intrinsically related to the two types of trading sys-
fourth quarter of 2007 alone (see TIME, February, 5, 2010). Deutche tems: the brokered market and the dealer market.3 In a brokered
Bank lost 4.8 billion euro in the fourth quarter, hurt by 1 billion euro in market, brokers present information in the form of the market price to
proprietary trading in credit market (see The Wall street Journal, potential buyers and sellers and collect transaction fees (equivalently,
February 6, 2009). set a bid-ask spread) in return. There are no special agents who trade
Therefore, the global financial crisis has triggered a reassessment of assets on their own account to make a market (i.e., proprietary trading
the economic costs and benefits of banks' involvement in proprietary is prohibited). On the other hand, in a dealer market, dealers play the
trading and other activities in financial markets. In response to the role of market makers and determine the price at which they are willing

*
Corresponding author.
E-mail addresses: [email protected] (K. Nishide), [email protected] (Y. Tian).
1
Certainly, there are many differences among these proposals, we here focus on the aspect of proprietary trading.
2
It seems that the industry and regulators have different definitions of proprietary trading. Here, we follow Gambacorta and van Rixtel (2013) as in Table 1 to
divide proprietary trading into two categories: “deal as principal in securities and derivatives” and “engage in market making”.
3
In the following, we use the term “proprietary trading” for “proprietary trading as market-making activities”. In the real world, investors trade financial assets in
mainly three types of market organizations: an auction market, a brokered market, and a dealer market. In an auction market, buyers and sellers directly confront
each other when bargaining their price. When there are insufficient participants in an auction market, investors need to trade through brokers (see, e.g., Ritter, Silber,
& Udell, 2008).

https://doi.org/10.1016/j.irfa.2019.101371
Received 31 March 2018; Received in revised form 31 May 2019; Accepted 4 July 2019
1057-5219/ © 2019 Elsevier Inc. All rights reserved.

Please cite this article as: Katsumasa Nishide and Yuan Tian, International Review of Financial Analysis,
https://doi.org/10.1016/j.irfa.2019.101371
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

Table 1 research questions: What is the impact of proprietary trading on (i) the
Comparison of selected structural reform proposals related to proprietary asset price and the transaction fee, (ii) the trading volume, and (iii)
trading. social welfare?
Source: Source: Gambacorta and van Rixtel (2013). The main contribution of this study is its effective demonstration of
Approach Volcker Liikanen Vickers how proprietary trading with transaction fees affects market equili-
brium solutions. It also provides several new testable implications for
Deal as principal in securities and derivatives No No No
empirical studies as follows.
Engage in market making Yes No No
Regarding question (i) above, we find that the bid and ask prices set
by the leader who can engage in proprietary trading are more favorable
to buy and sell an asset (i.e., proprietary trading is allowed). A natural price to investors, i.e., a lower ask price (for buy orders) and a higher
question arises: which system is better from the viewpoint of market bid price (for sell orders), even though the dealer monopolistically
liquidity and welfare? To answer this question, we need to examine the seeks his own profits.
impact of proprietary trading on the asset price, the transaction fee, the The intuition is as follows. Suppose that the dealer's subjective
trading volume, and welfare of investors. fundamental value of the asset is higher than the average investor's.
As for the two trading systems, several theoretical studies are re- Then, the dealer has an incentive to set the mid-price between the two
lated to the current study, including Röell (1990), Fishman and fundamental values to seek an expected profit by proprietary trading.
Longstaff (1992), Sarkar (1995), and Bernhardt and Taub (2010). The The price is also favorable to the average investor since the average
models presented in these studies have examined the impact of dual investor can earn an expected profit by placing a sell order.
trading (trading on the dealer/broker's own account in addition to ex- For question (ii), proprietary trading is also found to induce a larger
ecuting customers' orders). In particular, they focus on how the dealer trading volume and a higher transaction fee. This observation is quite
strategically uses the private information, extracted from orders, to intuitive, because the mid-price in a dealer market is more favorable to
make profits. The standard models assume a risk-neutral and perfectly investors than in a brokered market. The result on the transaction fees
competitive market maker, which is distinguished from a dealer. Under (equivalently, the bid-ask spread) is related to Colliard and Foucault
this assumption, the market price is equal to the conditional expecta- (2012), who show that reduced transaction fees is not necessarily
tion of the asset payoff, given the market maker's information, as in beneficial to investors, although the economic mechanism is different
Kyle (1985). However, in reality, in many cases, one agent plays the from ours.
role of the dealer and the market maker simultaneously. Moreover, in As to question (iii), we find that social welfare (the expected utilities
actual financial markets, competition among market makers is not of both the dealer and average investors) is improved. The dealer ob-
perfect. Although obvious conflicts of interest exist that are inherent in viously benefits from proprietary trading because he can set the price in
determining prices when dealers execute customers' orders against their addition to the transaction fee. The average expected utility of the in-
own account, market makers may have market power to determine or vestors also increases by virtue of a more favorable price, even the
at least to affect asset prices. Also, unlike many of the previous studies transaction fee is a little bit higher compared to that in a brokered
mentioned, the dealer/broker and the investors are not necessarily risk market.
neutral. The mechanism behind these results is as follows. Compared to the
Since existing models focus on informed trading, transaction fees in broker's final wealth, which consists only of the fee revenue, the dealer's
imperfectly competitive situations are rarely considered. However, as final wealth consists of two terms. In addition to the fee revenue, which
the literature of market microstructure grows, the effect of transaction always has a negative effect for investors, the dealer also earns a profit
fees on market participants becomes an important issue (e.g., Kyle & from proprietary trading, which can have a positive effect for investors,
Obizhaeva, 2013). Two exceptions are Colliard and Foucault (2012) depending on the realized value of the random payoff of the asset.
and Sarkar (1995). Colliard and Foucault (2012) develop a model of Proprietary trading enables the dealer to set a more favorable price for
limit order trading with trading fees and find that reduced trading fees investors to seek profits. Consequently, proprietary trading as market-
is not necessarily beneficial to investors. Sarkar (1995), which is more making activities is always beneficial to average investors.
related to our model, incorporates a commission fee as an extension to The result of higher transaction fees is consistent with Huang and
his basic dual trading model. In his study, the transaction fee is con- Stoll (1996), who reported that the cost of executing transactions is
sidered as a charge for market access, which is independent of the order higher on NASDAQ (dealer market) than on the NYSE (auction market)
size and dependent only on whether or not traders make a trade. The by every measure they calculated. The result of improved social welfare
dealer determines the fee by the zero-profit condition that his expected in a dealer market is consistent with Fishman and Longstaff (1992),
trading profits plus his income from the expected fee equal the total who examined dual trading in futures markets and empirically found
costs of the brokerage activities. Notably, the costs are exogenously that dual traders earn higher profits than non-dual traders, and that
given, and the commission fee is determined irrespective of the asset customers of dual-trading dealers do better than customers of non-dual-
price. trading brokers.
The objective of this study is to examine the impact of proprietary The remainder of this study is organized as follows. In Section 2, we
trading with transaction fees. Unlike in Sarkar (1995), the transaction set up our model. In Section 3, we solve and derive the equilibriums
fee is assumed to be proportional to the order amount. We construct a with and without proprietary trading. The numerical analysis is con-
one-period CARA-Normal model with a monopolistic dealer/broker and ducted in Section 4 to examine how proprietary trading with transac-
an infinite number of investors. The risk-averse investors in this study tion fees affects market equilibrium and investors' welfare. We discuss
have heterogeneous initial endowments and a heterogeneous belief in Section 5 the results of the current paper in comparison with those of
about the liquidation value of the asset. the previous studies. Section 6 presents the conclusions. Appendices
While the broker who brings the investors' orders to the market A–C provides the proofs of the main results.
(with no proprietary trading) sets the transaction fees based on the
market-clearing price, the dealer who trades on his own account (with 2. Model
proprietary trading) simultaneously determines both the transaction fee
and the asset price. This study attempts to address the following Consider a one-shot financial market in which two types of assets,
risk-free and risky assets, are traded. The two assets are traded among

2
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

agents at time 0. The risk-free asset plays the role of storage technology called CARA-Normal model due to mathematical tractability. In our
in that the interest rate is zero.4 The payoff of the risky asset, denoted model, all random variables follow normal distributions and all agents
by v, is random at time 0 and is realized at time 1. are assumed to have exponential utilities like Eq. (2.2). The max-
There are two types of market participants: investors who are price imization of Eq. (2.2) is equivalent to the maximization of
takers and a monopolistic agent (dealer or broker) who collects trans- a
action fees from investors and clears their orders. The difference be- [Yi (p , c )| i] [Yi (p, c )| i],
2 (2.3)
tween a dealer and a broker is that while the dealer trades with in-
vestors on his own account (i.e., with proprietary trading), the broker where γ is a risk-aversion coefficient. Then, the equilibrium solution is
only clear investors' orders (i.e., with no proprietary trading). easily derived as in Kyle (1989) and Kim and Verrecchia (1991).
Both the transaction fee c and the asset price p are determined at Since v is the only random variable in Eq. (2.1) and follows a
time 0. The transaction fee c is assumed to be proportional to the normal, the optimal trading volume of investor i is easily obtained by
trading amount; that is, investors have to pay c to either buy or sell a maximizing Ui(p,c) in Eq. (2.2). We thus have
unit of the asset.5 Therefore, p + c is regarded as the so-called ask price, µi p sgn[x i* (p, c )] c
while p − c is the bid price. x i* (p , c ) = i + ,
a 2
v (2.4)
Before we describe the model setup in details, we show the timeline
of our model as follows: At time 0, where the first term represents the hedging motivation for the risk in-
herent in the initial endowment, and the second term describes the
(i) investor i is endowed with a position ωi of the risky asset v. profit-seeking motivation.
(ii) investor i and dealer/broker have subjective belief v| i and v| M , Let i: =µi a v2 i , which can be interpreted as the risk-adjusted
based on their own information sets i and M , respectively, mean of investor i’s belief about the asset value. Then, Eq. (2.4) can be
(iii) dealer/broker sets the mid-price p and transaction fee c for the rewritten as
risky asset v, based on his own subjective belief as well as the (p + c ) (p c)
i i
anticipation of investor i’s order xi, x i* (p , c ) = 1{ i > p + c } + 1{ i < p c} .
a 2
a 2
(2.5)
(iv) investor i observes p and c, and submits the order xi. v v

(v) dealer/broker clears the orders from investors. Eq. (2.5) clarifies how investor i optimally trades the asset; that is, if
the parameter ζi is strictly higher (lower, respectively) than the ask
At time 1, the payoff of the risky asset v is realized. price p + c (the bid price p − c, respectively), then investor i buys (sells,
respectively) the asset. If p − c ≤ ζi ≤ p + c, then investor i does not
2.1. Investor's optimization problem trade the asset due to the presence of the transaction fee. Note that the
subscript i only appears in ζi. In other words, investors' heterogeneity is
We denote the set of investors by and index each investor by fully characterized by the parameter ζi.
i . Following Kim and Verrecchia (1991) and other extant studies, To simplify the problem and make it mathematically tractable, we
we suppose that investor i is endowed with an amount ωi of the risky assume that there are an infinite number of investors. More concretely,
asset and { i }i follows IID N ( ¯ , 2) . In addition, the payoff of the we let = R . Furthermore, μi, the mean of investor i’s subjective belief
risky asset is assumed to follow a normal distribution with respect to about the payoff v, is assumed to follow
the information of investor i, i.e., 2
µi N (µI , I) (2.6)
2
v| N (µi , v ), 7
i
in and be independent of {ωi}. In this situation, because of the in-
where i is the information set of investor i.6 The assumption that the dependence between μi and ωi, the parameter i = µi a v2 i follows a
heterogeneity lies in the expected value of v and the variance v2 is normal as
common and makes the problem simple. 2
N (µ , ), (2.7)
Let xi be the trading amount of the risky asset by investor i, where a i

positive (negative, respectively) value of xi indicates that investor i buys where µ : =µI a 2
v ¯ and 2
:= 2
I + a2 v4 2 8
.
(sells, respectively) |xi| units of the risky asset. Given the above as-
sumptions, the final wealth Yi is expressed as 2.2. Broker's or dealer's optimization problem
Yi (p, c ) = v i + (v p) x i sgn[xi ] cx i , (2.1)
Let M denote the information set of the dealer or broker. The
where sgn[x] = 1{x > 0}− 1{x < 0}. payoff of the risky asset v is also assumed to be a normal with respect to
Investor i’s utility is given by the exponential utility as the broker's/dealer's information set M , i.e.,
1
(2.8)
2
Ui (p, c ) = log ( [e aYi (p, c )
i]). v| N (µM , M ).
a (2.2) M

The dealer/broker's utility function is given by the exponential

For simplicity, the coefficient of absolute risk aversion, a, is assumed to
utility as
be common among investors.
Definition 1. (Investor's optimization problem). Given the dealer/ UM (p , c ) =
1
log ( [e R (p, c ) |
M ]),
broker's price and fee schedule (p,c), investor i determines his/her (2.9)
optimal trading strategy x i* (p , c ) by maximizing Eq. (2.2).
where R(p,c) is the dealer/broker's final wealth and γ represents the
In the market microstructure literature, many studies employ the so- coefficient of his absolute risk aversion.

4 7
Many studies in the market microstructure literature, such as Kim and The assumption (2.6) is justified by the central limit theorem if the sub-
Verrecchia (1991), assume a zero-interest rate. The results do not essentially jective mean is decomposed into a common term and an idiosyncratic IID term
differ if a non-zero interest rate is assumed. as in Hellwig (1980).
5 8
In the following, we use the words “proportional transaction fee” and Although investors and the dealer/broker are heterogeneous in beliefs and
“transaction fee per trade” interchangeably. endowments, they have full information on the distribution of ζi as N (µ , 2) ex
6
We do not explicitly consider the information structure { i}i . ante.

3
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

First, we define the optimization problem of the monopolistic into two terms: the amounts of total buy orders X+(p,c) and sell orders
broker. As explained before, the broker charges a proportional trans- X−(p,c), respectively.
action fee denoted by c, but is not allowed to trade the risky asset on his
own account. Hence, we can define the broker's optimization problem xi* (p , c ) = X+ (p , c ) + X (p , c ),
(3.1)
as follows: i

Definition 2. (Broker's optimization problem). Given investor i’s where

optimal trading strategy x i* (p , c ) in Eq. (2.9), the broker determines
± (p ± c )
X ± (p , c ) = ± qI ( )d ,
(i) the price p = pb by the market-clearing condition p±c a v2 (3.2)

xi* (p , c ) = 0; and
i (2.10)
( µ )2
2 2
(ii) and the fee c = cb by maximizing Eq. (2.9), where qI ( ) = e (3.3)

R (p , c ) = c × |xi* (p , c )|. is the density function of N (µ , 2

). We denote the trading volume
i (2.11) by X(p,c), as

X (p , c ) = |x i* (p , c )| = X+ (p , c ) X (p , c ).
i (3.4)
On the other hand, the dealer who plays the role of a market maker
determines the price at which he is willing to buy and sell the asset. The derivation of the two amounts (3.1) and (3.4) are provided in
Therefore, the dealer's optimization problem can be described as fol- Appendix A.
lows:
Definition 3. (Dealer's optimization problem). Given investor i’s
optimal trading strategy x i* (p , c ) , the dealer determines the price 3.1. A brokered market
p = pd and the fee c = cd by maximizing Eq. (2.9), where
First, we show the equilibrium solution for a brokered market. As
R (p , c ) = (v p) × ( x i* (p, c )) + c |xi* (p , c )|. shown in Appendix A, there is no randomness in the broker's optimi-
i i (2.12) zation problem. That is, the utility function UM(p,c) is equal to R(p,c),
which is defined in Eq. (2.11). Consequently, the problem is irrelevant
to the broker's risk aversion. By maximizing the fee revenue, we have
The first term of Eq. (2.12) represents the profit earned from pro- the following proposition.
prietary trading. Note that this term can be negative. In other words, Proposition 1. The equilibrium mid-price in a brokered market, pb, is given
depending on the realized value of the payoff v, the dealer may suffer by
losses from his or her proprietary trading. The second term describes
the fee revenue earned from investors. Unlike the first term, this term is pb = µ (3.5)
always positive.
and the proportional transaction fee by
It should be emphasized that the research focus of our paper is
different from that of Sarkar (1995), who mainly examined the impact cb = z , (3.6)
of dual trading (trading on the broker-dealer's own account in addition
to executing customers' orders) on both informed and uninformed tra- where z ∈ (−1,0) is the solution of the equation
ders. He also extended the model to include a commission fee that was
1 d
independent of the order size and dependent only on whether or not an z+ log (z ) = 0.
2 dz (3.7)
investor traded with the dealer/broker. In this study, we focus on how
proprietary trading as a market-making activity affects the asset price
and the proportional transaction fee; thus, the transaction fee is en-
dogenously determined in equilibrium. Proof. See Appendix A.
We should also mention that the differences in the research focus We have two findings from Proposition 1. First, with no proprietary
generate differences in the assumptions. In Sarkar (1995), to con- trading, the equilibrium price pb is set equal to the risk-adjusted mean
centrate on how the broker-dealer can make a profit from mimicking μζ of the investor's belief about the asset value v. It is a natural result
informed investors' trades, it is assumed that there is another risk- from market clearing in a brokered market, where the broker brings
neutral and perfectly competitive agent, the market maker, whose role investors' orders to the market without trading on his own account.
is to exclusively fix the asset price at which he will execute the total Second, we find that the transaction fee per trade cb depends on the
orders. On the other hand, we suppose a risk-averse and monopolistic standard deviation σζ of investors' belief. The more divergent the belief
broker (with no proprietary trading) or a dealer (with proprietary among investors, the higher the transaction fee per trade is set. In fact,
trading), each of whom simultaneously sets the market price of the the broker faces a trade-off between the trading volume i |x i* (p , c )|
asset and collects brokerage fees. Fig. 1 illustrates the difference be- in Eq. (2.11) and the fee revenue per trade c in determining the optimal
tween Sarkar (1995) and the current study in the model. transaction fee. If the broker sets a higher fee, although the fee revenue
per trade certainly is increased, we observe from Eq. (2.5) that the
3. Equilibrium solutions trading volume may be reduced due to an expanse of the no-trade re-
gion ζi ∈ (pb − cb,pb + cb), which may lead to a decrease in the total fee
In this section, we derive the market equilibrium solutions for the revenue. The optimal transaction fee cb in Eq. (3.6) is exactly the result
brokered and dealer makers, respectively. To this end, we need to first of considering the trade-off between the trading volume and the fee
prepare the net amount of orders i x i* (p, c ) from all investors as revenue per trade described above. When the belief among investors
well as the trading volume i |x i* (p , c )| that appear in the optimi- becomes more divergent, it is less necessary for the broker to worry
zation problems captured by Eqs. (2.10)–(2.12). about the decrease in trading activities. Thus, the effect of fee revenue
We divide the net amount of orders i x i* (p, c ) from all investors per trade dominates, and a higher proportional transaction fee is set.

4
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

Fig. 1. Illustration of the difference be-

tween Sarkar (1995) and the current study
in the model. Sarkar (1995) mainly ex-
amined the impact of dual trading on both
informed and uninformed traders. To con-
centrate on how the broker can make a
profit from mimicking informed investors'
trades, it is assumed that there is another
risk-neutral and perfectly competitive
agent, the market maker, whose role is to
exclusively fix the asset price p. The brokers
mediate the trade between the market
maker and investors to collect a transaction
fee c, which is assumed to be independent
of the order size and the price p. The dif-
ference between panels (a) and (b) is whe-
ther there exists dual trading or not. In
panel (b), the broker can submit an order d
on his own account in addition to submit-
ting customers' orders xi. On the other
hand, the research focus in this study is to
examine how proprietary trading as
market-making activities affects the asset
price and the proportional transaction fee.
We do not discriminate investors as in-
formed and uninformed. Instead, we as-
sume that a risk-averse (more general
compared to risk-neutral) and monopolistic
(observed market power to determine the
asset price) agent plays the role of the dealer/broker and the market maker simultaneously. Moreover, the transaction fee c|xi| is assumed to be proportional to the
trading amount xi and thus affected by the asset price p. While in panel (c), the broker who brings investors' orders to the market (with no proprietary trading) sets
only the proportional transaction fee c, given that the price p is determined according to the market-clearing condition among investors, in panel (d), the dealer who
trades on his own account (with proprietary trading) simultaneously sets both the asset price and the proportional transaction fee (p,c).

3.2. A dealer market After calculations, we have the equilibrium solution for a dealer
market as follows.
Now consider the optimization problem of the dealer. Similar to Eq. Proposition 2. The equilibrium mid-price and the proportional transaction
(2.3), the maximization of Eq. (2.9) is equivalent to the maximization of fee in an dealer market, pd and cd, are given by

[R (p, c )| M] [R (p, c )| M ].
(3.8) d^+ d^
2 pd = µ ,
2 (3.11)
By substituting R(p,c) defined in Eq. (2.12) into Eq. (3.8), we have the
equivalent objective function as d^+ + d^
cd = ,
2 (3.12)
(v p) ( x i*) + c |x i*|
whered^± are determined by the simultaneous equation system (B.5).
M M
i i

Proof. See Appendix B.

(v p) ( xi*) M ,
2 i (3.9) Intuitively, d^± in Eq. (B.6) are the risk premiums per unit of risk paid
by buy and sell orders, respectively. And d+ + d− = −2cd/σζ in Eq.
where the first term represents the expected profits from proprietary (B.7) is the twice of the transaction fee per unit of risk required by the
trading, the second term is the revenue from collecting transaction fees, dealer. Moreover, d+ + d− < 0 ensures that the proportional transac-
and the third term reflects the risk-aversion effect that stems from tion fee cd is positive. The sign of d+ − d− in Eq. (B.8) depends on the
proprietary trading, the profit from which is random. sign of μζ − pd, which can be positive or negative. Therefore, the
To clarify the intuition as well as the calculation, we further divide magnitude relation between pd and pb depends on the divergence be-
the first two expectations into terms for buy orders and sell orders, tween the belief of the dealer and the average investors. We will discuss
respectively. this point in details in Section 4.
In a dealer market, according to Eq. (3.9), the dealer should con-
(v p) ( x i*) M = (v p ) ( xi* ) M
sider three effects. The first effect is for the expected profit from pro-
i xi* > 0 prietary trading, which is absent in a brokered market. By setting a
proper mid-price, the dealer can earn a higher profit that the broker
cannot do. The second effect is the fee revenue, the same as in
+ (v p ) ( x i*) M ,
Proposition 1. Finally, the risk aversion of dealer brings a motivation to
xi* < 0
avoid inventory risk due to proprietary trading, which is also absent in
a brokered market. Eq. (B.5) is the condition that maximizes the dealer's
c |x i*| = c |x i*| + c |x i*| . utility, taking the three effects into account, where the first line ex-
M M M
i xi* > 0 x i* < 0 presses the marginal utility from setting a lower ask price pd + cd and a
higher bid price pd − cd, and the second line denotes the marginal
(3.10) disutility due to risk aversion.

5
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

Although analytical solutions seem difficult to obtain, we conduct Table 2

numerical analysis in the next section to analyze how the two markets Summary of key variables and parameters.
differ, especially from the viewpoint of investors' welfare. Investors
¯ Mean of initial endowment 1
3.3. Relationship between the two markets σω Standard deviation of initial endowment 1
σv Standard deviation of belief about asset value 0.5
μI Mean of average belief about asset value 1
We find an important analytical property that connects the two σI Standard deviation of average belief about asset value 1
markets. a Coefficient of absolute risk aversion 1
Corollary 1. The market equilibrium in the limiting case of an infinite risk- Broker/dealer
averse dealer is equal to the one in the broker's case; that is, μM Mean of belief about asset value 1
σM Standard deviation of belief about asset value 0.25
pd pb γ Coefficient of absolute risk aversion 0.5
.
cd cb

proportional transaction fee is lower. However, with the figures below,

Proof. See Appendix C. we will find that the opposite is true: on average, a dealer market is
more favorable to investors.
Intuitively, when γ →∞, which means the dealer is extremely risk- Second, we analyze the impact of μI on p, the asset price. Panel (b) of
averse, he would never take any risk and would, thus, avoid any pro- Fig. 2 shows that the asset price increases with μI. Intuitively, the higher
prietary trading. Since the profit from transaction fees is not random the mean of the average investors' belief μI about the asset value, the
and is always positive, maximizing the dealer's wealth is equivalent to higher the price p is set, because investors are anticipated to buy the
maximizing the fee revenue, leading to the same equilibrium as is asset. Moreover, we find that pd < pb (the asset price set by the dealer is
achieved in a brokered market. lower than the one set by the broker) for μζ > μM (on the right side of
the intersection of the solid blue and red lines), and pd > pb for μζ < μM
4. Numerical analysis (on the left side of the intersection). In other words, the slope of the
asset price with proprietary trading is more moderate compared to the
This section conducts numerical analysis to investigate the effect of price with no proprietary trading. Although the proportional transac-
proprietary trading on equilibrium. The base case parameter values are tion fee is set a little bit higher in a dealer market, the ask and bid prices
given in Table 2. Note that we have μζ < μM in this parameter setting, including transaction fees are favorable to investors. The reason can be
implying that the risk-adjusted mean of the average investors' belief is interpreted as follows: The larger the divergence between the beliefs of
lower than that of the dealer's/broker's belief. the dealer and average investors, the active the trading activities are,
and the higher the revenue from transaction fees will be. Therefore, it is
4.1. Impact of mean of average investors' belief not necessary for the dealer to adjust the price drastically.
Third, we consider the impact of μI on the trading volume X(p,c)
First, we examine the impact of μI, the average expectation about defined by Eq. (3.4). Panel (c) of Fig. 2 indicates that the trading vo-
the asset value v among investors. The parameter μI affects the equili- lume in the two markets coincides at µI = µM + a v2 ¯ , or equivalently
brium only through μζ, which is equal to µI a v2 ¯ . The average initial μζ = μM. The larger the divergence between the belief of the dealer and
endowment among investors, ¯ , also appears only in μζ. Since it is average investors, the higher the trading volume in a dealer market is,
obvious that ¯ has an opposite effect compared to μI, we omit the since proprietary trading enables investors to trade more actively to
analysis of ¯ . seek profits. Consider the case μζ > μM. Average investors do not trade
Fig. 2 depicts the effect of μI on the equilibrium solutions. in a brokered market due to the presence of the transaction fee (μζ ∈
First, we find from panel (a) of Fig. 2 that the proportional trans- (pb − cb,pb + cb)). In a dealer market, however, average investors may
action fee in a brokered market is constant, regardless of μI, reconciling trade, depending on the relationship between μζ and pd ± cd. As we
result (3.6) in Proposition 1. have seen from panel (b), when μζ > μM, the price in a dealer market is
Intuitively, it is not necessary for the broker to take into con- more favorable to the average investors, and they are willing to buy the
sideration investors' average expectation about the asset value and asset if their risk-adjusted belief μζ is higher than the ask price pd + cd.
compare his own expectation μM with μI, since the broker does not Fourth, we examine the impact of μI on investors' welfare UI in Panel
engage in proprietary trading. (d) of Fig. 2. Here, the welfare of the total investors is naturally defined
On the other hand, the transaction fee set by the dealer varies, de- by
pending on μI, more exactly, depending on the divergence between the
UI = Ui ( ) qI ( )d . (4.1)
dealer's own expectation μM and investors' expectation μI. A key ob-
servation is that the graph for cd is of U-shape and the minimum is
From panel (d), we find that the welfare of the total investors is
cd = cb, achieved at µI = µM + a v2 ¯ , or equivalently μζ = μM.9 Recall
higher in a dealer market. An intuitive explanation for this result is
that μζ is the risk-adjusted mean of average investors' belief about the
similar to that given for panel (c); that is, the trading price is more
asset value. Intuitively, if the belief of both the dealer and average in-
favorable to average investors, because pd + cd < pb + cb when
vestors coincides, the result in the two markets is the same and is in-
μζ > pd + cd (average investors are willing to buy the asset) and
dependent of proprietary trading.
pd − cd > pb − cb when μζ < pd + cd (average investors are willing to
The result of a higher transaction fee in the dealer market is con-
sell the asset). Therefore, their expected profit becomes higher.
sistent with Huang and Stoll (1996), who reported that the cost of ex-
Fifth, we study the effect of μI on the utility UM of the dealer/broker.
ecuting transactions is higher on NASDAQ (dealer market) than on the
Panel (e) of Fig. 2 shows that the utility of the dealer is always higher
NYSE (auction market) by every measure they calculated. At first
than that of the broker, except for the case where µI = µM + a v2 ¯ :
glance, a brokered market is more favorable to investors because the
here, the utility coincides in the two markets. The reason for this result
can be intuitively explained from the fact that the dealer has two
9
Since the first line of Eq. (B.5) disappears when μζ = μM, the first-order control variables p and c, while the broker can only optimally determine
condition of the dealer's maximization problem reduces to Eq. (A.3). Therefore, the proportional transaction fee c.
we have cd = cb. Panels (d) and (e) of Fig. 2 indicate that although the dealer has

6
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

Fig. 2. The impact of mean μI of average investor's belief on (a) proportional transaction fee c, (b) asset price p, (c) trading volume X, (d) investors' welfare UI, (e)
utility of the dealer/broker UM, respectively.

monopolistic power to set the price and the transaction fee, on average, equilibrium only through σζ, which is equal to I + a2 v4 2 . The devia-
a dealer market is desirable for investors. This observation of improved tion in initial endowment among investors, σω, also appears only in σζ.
social welfare in a dealer market is new in the literature, especially in Since it is obvious that σω has a similar effect with σI, we omit the
studies that compare the two major trading systems in actual financial analysis of σω.
markets. Moreover, the result is consistent with Fishman and Longstaff Fig. 3 shows how this deviation of belief affects the equilibrium
(1992), who examined dual trading in futures markets and empirically solutions.
found that dual traders earn higher profits than non-dual traders, and First, we find from panel (a) of Fig. 3 that the transaction fee is an
that customers of dual-trading dealers do better than customers of non- increasing function of the deviation of investors' belief in both cases:
dual-trading brokers. that is, the more divergent the belief among investors, the higher the
transaction fee is set. Also note that the fee per trade in a dealer market
4.2. Impact of deviation in average investors' belief is always a little bit higher than that in a brokered market. However, we
should consider not only the transaction fee but also the price as in the
In this analysis, we investigate the impact of σI, the standard de- analysis of μI.
viation of investors' subjective belief {µi }i . The parameter σI affects the Second, panel (b) of Fig. 3 describes the impact of the deviation in

7
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

Fig. 3. The impact of standard deviation σI in average investor's belief on (a) proportional transaction fee c, (b) asset price p, (c) trading volume X, (d) investors'
welfare UI, (e) utility of the dealer/broker UM, respectively.

average investors' belief on price p in the two markets. It is worth trading volume is higher in a dealer market than in a brokered market
mentioning that pb < pd in the panel. That is, in our basic parameter because the price is more favorable to average investors, leading to
setting where μζ < μM, we have μζ = pb < pd. Thus, as in panel (b) of more active trading activities by investors.
Fig. 2, the price is more desirable to average investors in a dealer Fourth, panel (d) of Fig. 3 plots the impact on investors' welfare UI.
market than in a brokered market, because investors are anticipated to We observe that a higher value of σI leads to a higher value of welfare.
sell the asset. As seen in panels (a)–(c) of Fig. 3, more investors are motivated to trade
Third, we examine the impact on X(p,c), the total trading volume in the asset when σI is high, because the risk-adjusted mean and the price
the market. As seen from panel (c) of Fig. 3, the trading volume substantially differ. The divergence between μζ and p results in an in-
monotonically increases with the deviation in belief. An economic ex- crease in welfare. Moreover, welfare in a dealer market is higher than in
planation is given in a similar way to that in panel (a) of Fig. 3 as a brokered market because the price is more favorable to average in-
follows: Suppose that σI becomes higher. Then, for more investors, the vestors, as shown in the previous analyses.
risk-adjusted mean is far from the price, or ζi∉(p − c,p + c). Conse- Lastly, we present panel (e) of Fig. 3 to describe how the deviation
quently, they are more willing to trade the risky asset. Again, the in belief affects the utility UM of the dealer/broker. As with investors'

8
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

Fig. 4. The impact of investors' degree of risk aversion a on (a) proportional transaction fee c, (b) asset price p; and the impact of dealer/broker's degree of risk
aversion γ on (c) proportional transaction fee c, (d) asset price p, respectively.

welfare, the utility of the dealer/broker is increasing in σI. We omit a 5. Discussions

detailed explanation of the figure because it is simply a restatement of
that for panel (e) of Fig. 2. From the numerical results above, we can summarize that a dealer
market is more desirable to investors compared to a brokered market,
even though the dealer has a monopolistic power to set the asset price
4.3. Impact of risk aversion and the transaction fee. In other words, proprietary trading as market-
making activities leads to a favorable consequence for financial mar-
The impact of the risk-aversion coefficients of investors and the kets.
dealer/broker are described in Fig. 4. The economic intuition is as follows: Recall that the dealer's profit is
Panels (a) and (b) of Fig. 4 depict the effect of a, the risk aversion given by Eq. (2.12), the first term of which expresses the profit from
coefficient of investors. As investors become more risk averse, their proprietary trading. A more favorable ask price p + c and bid price
demand for risk hedging grows. Thus, the dealer is able to set a higher p − c for investors attracts more orders, leading to a higher value of the
proportional transaction fee. Since μζ < μM in this case, sell orders in- trading volume X(p,c). Hence, even though the dealer is monopolistic
crease, the price is set lower. Moreover, the dealer set a more favorable and risk-averse in our model, he has an incentive to set a more favor-
price for investors. able price for investors to earn a higher profit from proprietary trading.
In parallel, panels (c) and (d) of Fig. 4 describe the effect of γ, the In a brokered market, the broker always set the price pb = μζ to clear
risk aversion coefficient of the dealer/broker. When the dealer becomes the investors' orders. Consequently, investors with ζi ∈ (pb − cb,pb + cb)
more risk averse, he is not willing to hold the asset. Thus, if the total net do not trade the asset because the proportional transaction fee cb dis-
supply of the asset in the market is positive, both the price and the courages them from doing so. Put differently, investors whose belief is
transaction fee are set lower to induce investors to buy the asset. close to the average belief in the market always decide not to trade in a
However, the risk aversion of the broker does not affect the equilibrium brokered market.
due to the absent randomness in the broker's optimization problem (see In a dealer market, things are different. Suppose μζ > μM, i.e., the
the details aforementioned before Proposition 1). As we proved in risk-adjusted mean of average investors' belief about the asset value is
Corollary 1, when the dealer is infinitely risk-averse (γ →∞), the higher than the dealer's belief. In this situation, the dealer has an in-
equilibrium solution (pd,cd) in a dealer market converge to (pb,cb) in a centive to set the ask price pd + cd lower since it can attract buy orders
brokered market. and make the dealer earn more profits from proprietary trading. On the
The effects of other parameters are also obtained by numerical other hand, a lower ask price is also more favorable to average investors
calculations. Since the qualitative results and the economic explana- because they are willing to buy the asset if pd + cd < μζ. The same si-
tions are similar to those in the previous cases, we omit the illustrations, tuation applies to the case where μζ < μM. Consequently, a dealer
which are available from the authors on request. market is, on the whole, better for investors than a brokered market

9
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

Table 3 price and fee is not fully investigated in the literature. To the best of our
Comparison between our results with those of Sarkar (1995). Here, “d.t.” and knowledge, the present study is the first to investigate an equilibrium in
“p.t.” represent dual trading and proprietary trading, respectively. which the market price and transaction fee are set by one agent under
Sarkar (1995) Our study the condition of imperfect competition.

Transaction fee Lower with d.t. Higher (per trade) with p.t. 6. Conclusion
Trading volume Lower with d.t. Higher with p.t.
Welfare Lower with d.t. for informed Higher with p.t. on average
Higher with d.t. for uninformed In this study, we construct a one-period model in which a market
arranger and an infinite number of investors participate in trading a
risky asset. While the broker who brings investors' orders to the market
thanks to proprietary trading. (with no proprietary trading) sets only the transaction fee, given that
Now suppose that γ is large; equivalently, the dealer is quite risk the price is determined according to the market-clearing condition
averse. The dealer is unwilling to take the risk of proprietary trading among investors, the dealer who trades on his own account (with
but is willing to earn profits from the second term of Eq. (2.12), fee proprietary trading) simultaneously sets both the transaction fee and
revenues, because this revenue source is deterministic. In this case, the the asset price. We find that proprietary trading enables the dealer to
equilibrium of the limiting case converges to that in a brokered market. set a more favorable price not only for the dealer himself but also for
Our results are related to Colliard and Foucault (2012), who con- investors. As a result, the trading volume and the transaction fee per
struct a limit-order market model and show that a lower transaction fee trade both increase, and social welfare improves.
does not necessarily imply improvement of market quality, because a Finally, we should point out an interesting but challenging question.
lower transaction fee can induce investors to post-limit orders with a Our model could be extended to an intermediate oligopolistic case
smaller execution probability. In our model, although the transaction between the extremes of a perfectly competitive market arranger in the
fee per trade is higher in a dealer market, the mid-price is more fa- standard models and monopolistic one in our model. The study of such
vorable to the average investors, and so do the bid and ask prices. an oligopolistic equilibrium can be taken up in future research. While
Therefore, the trading volume increases and the average investors' the focus of this paper is proprietary trading as market-making activ-
utility gets higher. ities with transaction fees, we are silent on proprietary trading as
To further clarify the contribution of this study, we present Table 3 speculation and manipulation activities. To examine this interesting but
that briefly compares the results of our model with those in Sarkar challenging question, we need to incorporate detailed information
(1995), who studied the impact of dual trading with brokerage fees. structure into the model and consider the problem in a multiple-period
In Sarkar (1995), the commission fee is chosen by the condition of or continuous setting.
the broker's zero profit condition, or that the broker's expected trading
profit plus expected fee revenue equals the total costs of brokerage. Funding
Since the broker makes a positive profit from mimicking the informed
investors' trades, the commission fee is lower with dual trading. The This work is an output of one of the Project Research at the Institute
trading volume of informed investors as well as their welfare decreases. of Economic Research, Kyoto University (KIER) as the Joint Usage and
On the other hand, the welfare of the uninformed investor increases Research Center. The financial supports of the Japanese Ministry of
because their trading volume is unaffected and the commission fee is Education, Culture, Sports, Science and Technology (MEXT) Grant in
lower. Aid for Scientific Research (A) #25245046, (B) #16K17151, and (C)
To focus on the mimic behavior of the broker with dual trading, #26380390 are gratefully acknowledged.
Sarkar (1995) assumed that the price is set by a perfectly competitive
market maker, indicating that the price has no effect on the transaction Acknowledgments
fee. In contrast, our research interest is in proprietary trading as
market-making activities wherein a dealer/broker has monopolistic We thank the project members Masaaki Fukasawa, Chiaki Hara,
power over the asset price and the fee per trade. Our new findings are Masaaki Kijima, and Akihisa Tamura for their insightful comments and
that, since proprietary trading enables the dealer to set a more favor- suggestions. We also thank Durrel Duffie, Rohit Rahi, and seminar
able price for investors, trading volume and the transaction fee per participants at Kyoto University, Osaka University, Okayama
trade increase, and the welfare of average investors is improved. The University, and the 5th Paris Financial Management Conference for
effect of proprietary trading by a monopolistic dealer who sets both the their useful comments.

Appendix A. Proof for Proposition 1

Under the assumption (2.6), the net amount of orders from all investors is not random and is given by

x i* (p, c )= X+ (p , c ) + X (p , c )
i
(p + c ) p c (p c )
= qI ( )d + qI ( )d
p+c a v2 a v2
= 2
[( (d+) + d+ (d+)) ( (d ) + d (d ))],
a v (A.1)
where Φ and ϕ are the distribution and density functions of a standard normal, respectively, and
µ (p ± c )
d±: =± .

We also derive the trading volume, denoted by X, as

10
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

X (p , c )= |x i* (p , c )| = X+ (p, c ) X (p , c )
i

= 2
[( (d+) + d+ (d+)) + ( (d ) + d (d ))].
a v (A.2)
Neither Eqs. (A.1) or (A.2) is random, thus facilitating the mathematical tractability of the problem.
From Eq. (A.1), the market-clearing condition (2.10) becomes
( (d+) + d+ (d+)) ( (d ) + d (d )) = 0. (A.3)
It is easily verified that the price pb is equal to μζ. We should also notice that there is no randomness in the broker's optimization problem. By
substituting pb = μζ into the dealer's utility function (2.9), we have
2
UM (p , c ) = R (p , c ) = (d¯ (d¯) + d¯2 (d¯)),
a 2
v (A.4)
where d¯ = c/ .
The maximization of Eq. (A.4) is equivalent to the minimization of

R¯ (d¯): =d¯ (d¯) + d¯2 (d¯) (A.5)

with respect to d̄ . The first-order condition is given by

R¯ (d¯) = (d¯) d¯2 (d¯) + 2d¯ (d¯) + d¯2 (d¯) = (d¯) + 2d¯ (d¯) = 0,
where we have used the fact that (d¯) = d¯ (d¯) . Since R′(0) > 0 and R′(−1) < 0, we have at least one root of the equation R (d¯) = 0 for
d̄ ( 1, 0) . On the other hand, Eq. (3.7) is calculated as
d (z )
dz (z ) ( z ) + 2z ( z )
z+ =z+ = = 0,
2 (z ) 2 (d¯) (z )
indicating that d̄ in equilibrium must satisfy Eq. (3.7).
For the second-order condition, note that

R¯ (d¯) = d¯ (d¯) + 2 (d¯) = 2(1 d¯2) (d¯) > 0.

The inequality means that the minimizer of R̄ should be d̄ (0, 1) and unique, which completes the proof.

Appendix B. Proof for Proposition 2

Given p and c, the final wealth of the dealer is obtained by substituting Eqs. (A.1) and (A.2) into Eq. (2.12):

R (p , c )= (v p) ( x i* (p , c )) + c |xi* (p , c )|
i i

= (v p c) 2
[ (d+) + d+ (d+)]
a v

+ (v p + c) 2
[ (d ) + d (d )].
a v (B.1)
From the fact that v is the only random variable in Eq. (B.1) and follows a normal as Eq. (2.8), we can calculate Eq. (2.9) as

UM (p , c )= 2
(µM µ )[( (d+) + d+ (d+)) ( (d ) + d (d ))]
a v
2

2
[(d+ (d+) + d+2 (d+)) + (d (d ) + d 2 (d ))]
a v
2 2
M
[( (d+) + d+ (d+)) ( (d ) + d (d ))]2 .
2 a2 v4 (B.2)
Here, the first term represents the expected profit from proprietary trading as

a 2
(µ µM )( (d+) + d+ (d+))= (v p ) ( )
x i* M ,
v x i* > 0

a 2
(µ µM )( (d ) + d (d ))= (v p ) ( )
x i* M .
v x i* < 0

The second term represents the total revenue due to the collection of transaction fees.

2
(d+ (d+) + d+2 (d+)) = c |x i*| M ,
a v xi* > 0 (B.3)

11
K. Nishide and Y. Tian International Review of Financial Analysis xxx (xxxx) xxxx

2
(d (d ) + d 2 (d )) = c |x i*| M .
a v x i* < 0 (B.4)
Note that d should be negative as in Proposition 1. Finally, the last term of Eq. (B.2) reflects the risk-aversion effect that stems from the
proprietary trading, the profit from which is random.
2 2
M
[( (d+) + d+ (d+)) ( (d ) + d (d ))]2
a v4
2

= (v p) ( x i*) M .
i

Dividing Eq. (B.2) by / a v2 , and differentiating it with respect to d+ and d−, respectively, we obtain

(µ µM ) (d^±) [ (d^±) + 2d^± (d^±)]

2
(d^±)[( (d^+) + d^+ (d^+)) ( (d^ ) + d^ (d^ ))] = 0,
M
2
a v (B.5)
where
µ (pd ± cd)
d^±: =± .
(B.6)
Note that
µ (pd + cd) µ (pd cd ) 2c
d^+ + d^ = = < 0,
(B.7)
and
µ (pd + cd) µ (pd cd ) 2(µ pd )
d^+ d^ = + = .
(B.8)
Therefore, we finally obtain

d^+ d^
pd = µ ,
2
d^+ + d^
cd= .
2

Appendix C. Proof for Corollary1

If γ →∞, then the variance of R(p,c) must be zero. Since v is the only random variable in Eq. (B.1), the part of i ( xi* (p , c )) must be zero,
which is exactly the market-clearing condition in a brokered market. Moreover, R(p,c) is reduced from Eq. (2.12) to Eq. (2.11), which is exactly the
final wealth of the broker. Therefore, we have pd = pb and cd = cb.

References 313–357.
Kim, O., & Verrecchia, R. E. (1991). Trading volume and price reaction to public an-
nouncements. Journal of Accounting Research, 29, 302–321.
Bernhardt, D., & Taub, B. (2010). How and when is dual trading irrelevant? Journal of Kyle, A. S. (1985). Continuous auctions and insider trading. Econometrica, 53, 1315–1335.
Financial Markets, 13, 295–320. Kyle, A. S. (1989). Informed speculation with imperfect competition. Review of Economic
Colliard, J. E., & Foucault, T. (2012). Trading fees and efficiency in limit order markets. Studies, 56, 317–355.
Review of Financial Studies, 25, 3389–3421. Kyle, A. S., & Obizhaeva, A. A. (2013). Market microstructure invariance: Theory and
Fishman, M. J., & Longstaff, F. A. (1992). Dual trading in futures markets. Journal of empirical tests. Working paper.
Finance, 47, 643–671. Ritter, L. S., Silber, W. L., & Udell, G. F. (2008). Principles of money, banking and financial
Gambacorta, L., & van Rixtel, A. (2013). Structural bank regulation initiatives: markets. New Jersey: Prentice Hall.
Approaches and implications. BIS Working Paper No. 412. Röell, A. (1990). Dual-capacity trading and the quality of the market. Journal of Financial
Hellwig, M. F. (1980). On the aggregation of information in competitive markets. Journal Intermediation, 1, 105–124.
of Economic Theory, 22, 477–498. Sarkar, A. (1995). Dual trading: Winners, losers, and market impact. Journal of Financial
Huang, R. D., & Stoll, H. R. (1996). Dealer versus auction markets: A paired comparison of Intermediation, 4, 77–93.
execution costs on NASDAQ and the NYSE. Journal of Financial Economics, 41,

12