Economics of Financial Markets

Market Microstructure, Final Exam

Prof. Giovanni Cespa

The exam is open book, and you have 24-hour time to return it. You have to answer all
questions.

1. Consider the Kyle (1985) model, but assume that instead of a single informed trader there
are N > 1 informed traders, who all perfectly observe the final value of the security v but
not the equilibrium price at the time that they determine their quantity demanded qi .

(a) Suppose that market makers post the Pprice schedule described by equation p(q) =
µ + λq, where q is the net order flow Ni=1 xi + u and µ = E[v]. Assuming that each
informed trader uses the following order submission strategy:

xi = Xi (v) = β(v − µ) for i ∈ {1, .., N },

find the value of β for which a Nash equilibrium exists, determine how β is affected
by N , and explain intuitively why.
(b) Suppose now that investors follow the order submission strategy derived in step (a).
Show that in this case the market makers’ pricing strategy is given by equation
p(q) = µ + λq, and find the value of λ that they optimally choose.
(c) What is the market depth in equilibrium, and how is it affected by an increase in
the number of informed traders, N ? What is the economic intuition for this result?

2. Consider a modified version of the static model by Kyle (1985) where market makers
conjecture that the informed trader, in equilibrium, may have the incentive to include a
random component in his order, thus increasing the noise in the market. Hence, their
conjecture about the trading strategy of the insider is:

x = β(v − µ) + φ,

where φ is normally distributed with mean zero and variance σφ2 , (that is φ ∼ N (0, σφ2 )
and it is such that Cov(φ, x) = Cov(φ, u) = Cov(φ, v) = 0).

(a) Assume that competitive market makers post the following price schedule:

p(q) = µ + λq,

where q = x + u is the net order flow. Find the competitive value of λ. Is market
depth higher or lower when compared to the baseline Kyle model? Explain intuitively
why.
(b) Solve the maximization problem of the informed trader, that is the value of x that
solves
max E [x(v − p)|v]
x

when the informed trader conjecture that the price schedule is p = µ + λq. What is
the optimal value of φ, in equilibrium? Explain your answer.

1
3. Consider the static model by Kyle (1985), where (i) market makers are risk neutral and
perfectly competitive, (ii) the asset value is v ∼ N (µ, σv2 ), (iii) the informed investor’s
order is x = β(v − µ), and the noise traders’ order is u ∼ N (0, σu2 ), independent of v;
and (iv) market makers only observe the total net order q = x + u. Suppose that the
insider is risk-averse (with constant coefficient of absolute risk aversion b > 0) and that
he liquidates any amount of the security that he buys at a liquidation value v + , where
 ∼ N (0, σ2 ), independent of u and v. At the time of trading the informed investor knows
the realization of v but not that of . This noise in his signal implies that in taking long
or short positions based on his privileged information, he bears some risk.

(a) Write down the mean-variance objective function of the insider, and obtain his de-
mand function from the solution of the first order condition associated to the insider’s
expected utility maximization problem. Identify the insider’s aggressiveness β RA .
(b) Execution risk is the risk that a transaction will execute at a price that is very far
away from recent market prices. Denoting by β K the insider’s trading aggressiveness
in the baseline Kyle (1985) case, show that β RA < β K (you can assume that the
equilibrium value of the market impact is positive: λ > 0). What is the intuition for
this result? How does it relate to execution risk? What happens to market liquidity?

4. In the lecture, we have investigated the situation in which order-processing costs are
assumed to be γ per share traded. Consider the following alternative assumptions:

(a) Assume that order-processing costs are k per transaction. Compute the bid-ask
spread in this case and show that it is decreasing with the size of the transaction.
Which features of the technology of trading would lead you to think that this is a
realistic model of order-processing costs?
(b) Assume that order-processing costs are k per euro traded. Show that the absolute
bid-ask spread is increasing in the security’s underlying value and the relative bid-
ask spread is constant, in contrast with the expressions found in the lecture where
order-processing costs are a constant γ per euro traded, irrespective of the share
value.

2
 Solutions
1. Kyle model with multiple informed traders.

(a) Informed trader i chooses a quantity xi that maximizes his expected profit:

max E[xi (v − p)]

xi

where he realizes that his quantity demanded

PN will affect the equilibrium price, since
it is determined by the total demand q = j=1 xj +u expressed by himself, his N −1
peers and the uninformed traders:

p = µ + λ [xi + (N − 1) β (v − µ) + u] .

Observing that E[u] = 0 and substituting this expression for the expected market
price into the maximand:

max xi [(v − µ)(1 − λβ(N − 1)) − λxi ]

xi

and taking the first-order condition with respect to xi yields

1 − λβ (N − 1)
xi = (v − µ).
2λ
Equating coefficients with the informed traders’ assumed order submission strategy:

1 − λβ (N − 1)
β=
2λ
yields the desired result. As usual in Cournot Nash competition, the greater the
number of competitors the less each one demands, though the sum of their demands
is increasing in N .
(b) The competitive market makers set the price equal to the expected value of the
security, given the demand from both informed and noise traders:
 
N
p = µ + E (v − µ)|q = u + (v − µ)
(N + 1)λ
N
σ2
(N +1)λ v
= µ+ N
q
( (N +1)λ )2 σv2 + σu2

Equating the slope coefficient to the posited value λ:

N √
σ2
(N +1)λ v N σv
λ= 2 =⇒ λ = .
N + 1 σu

N
(N +1)λ
σv2 + σu2

(c) This expression is decreasing in N ≥ 1. The more insiders compete with each other,
the deeper the market, as λ decreases.

2. Kyle model with a noisy insider’s order.

3
 (a) The conjectured trading by the informed trader is x = β(v − µ) + φ, so that the
net aggregate order flow is q = β(v − µ) + u + φ. Hence market makers’ optimal
inference is
cov(v, q) βσv2
E[v|q] = µ + q =µ+ 2 2 q.
var(q) β σv + σu2 + σφ2
The zero-profit condition E[v|q] = p then implies

βσv2
p=µ+ q.
β 2 σv2 + σu2 + σφ2
| {z }
λ

Hence market depth is:

1 β 2 σv2 + σu2 + σφ2
= .
λ βσv2
Hence, we see that market depth is higher than in the baseline Kyle model. There-
fore, by creating some noise, the informed trader can actually reduce the price impact
of his order.
(b) As in the standard model by Kyle, the informed trader will choose x so as to

maxE [x(v − p)|v] = xE [(v − p)|v]

x

whose first-order condition is

v − µ = 2λx,
that is,
1
x= (v − µ) = β(v − µ)
2λ
Clearly, the optimal value of φ is zero, in equilibrium, since it can’t be optimal for
the informed investor to add a noise component to his order, thus deviating from
the profit maximizing order, even if this increases the liquidity in the market.

3. Kyle with a risk-averse insiders.

(a) The informed trader knows v but is uncertain about  as well as about the (unrelated)
impact λu of the noise trades on the market price. Assuming a linear price p =
µ + λ(x + u), due to CARA and normality, his objective function is therefore given
by:
b b
E[x(v +  − p)] − Var[x(v +  − p)] = x(v − µ − λx) − x2 (σ2 + λ2 σu2 ).
2 2
Differentiating the above function with respect to x, equating the derivative to zero
and solving for the insider’s strategy yields:
1
x= (v − µ).
2λ + b(σ2 + λ2 σu2 )
| {z }
=β RA

(b) The insider’s aggressiveness in the baseline Kyle model is given by β K = 1/2λ, and
the result follows. Note that both execution risk and final valuation risk reduce the
insider’s willingness to trade, given the market illiquidity equilibrium coefficient λ.

4
4. Alternative assumptions on order processing costs.

(a) If order-processing costs are k per transaction, they are k/q per share traded. Hence,
just replace γ = k/q in the expression for the spread, to obtain

k
St = 2 + sat + sbt ,
q

so that, in the absence of adverse selection (sat = sbt = 0), the bid-ask spread is
inversely proportional to shares traded. Fixed costs per transaction can arise if
the trading technology requires physical delivery of securities or paperwork for each
trade. Electronic trading has reduced these fixed costs per transaction.
(b) If order-processing costs are k per euro traded, they amount to kp per share traded,
so that the transaction price is given by:
µt
pt = µt + kpt dt =⇒ pt = ,
1 − kdt
where dt ∈ {−1, +1}. Hence, the bid and ask are given by:
µt µt
at = , bt = ,
1−k 1+k
and the bid-ask spread is
2kµt
St = ,
1 − k2
so that in this case the bid-ask spread is increasing in the security’s underlying value.
The relative bid-ask spread is instead constant:
St 2k
≡ st = .
µt 1 − k2