Strategic Insider Trading Equilibrium with a

non-fiduciary market maker

arXiv:1908.08777v1 [q-fin.TR] 23 Aug 2019

Knut K. Aase1 , and Bernt Øksendal2

[email protected], [email protected]
22 august 2019

Abstract
The continuous-time version of Kyle’s (1985) model is studied, in
which market makers are not fiduciaries. They have some market
power which they utilize to set the price to their advantage, resulting
in positive expected profits. This has several implications for the
equilibrium, the most important being that by setting a modest fee
conditional of the order flow, the market maker is able to obtain a
profit of the order of magnitude, and even better than, a perfectly
informed insider. Our model also indicates why speculative prices are
more volatile than predicted by fundamentals.
KEYWORDS: Insider trading, asymmetric information, strategic
trade, price distortion, non-fiduciary market maker, bid-ask spread,
linear filtering theory, innovation equation
Mathematics Subject Classification 2010: 60G35, 62M20, 93E10,
94Axx

1 Introduction
In his seminal paper on insider trading, Albert Kyle (1985) asks several
questions: How valuable is private information to an insider? How does
1
Norwegian School of Economics (NHH), Helleveien 30, N–5045 Bergen, Norway.
2
Dept of Mathematics, University of Oslo, P.O. Box 1053 Blindern, N–0316 Oslo, Nor-
way.

1
noise trading affect the volatility of prices? What determines the liquidity of
a speculative market? He provides answers to these and other questions by
modeling rigorously the trading strategy of an insider in a model of efficient
price information.
One important feature of a real securities market that remained unex-
plained in Kyle’s analysis is the existence of a bid-ask spread. Kyle focuses
on a single auction model in which a risky asset is exchanged for a riskless
asset among three kinds of traders. A single insider has access to perfect,
private observation of the ex post liquidation value of the risky asset. Un-
informed noise traders trade randomly. Market makers set prices and clear
the markets after observing the quantities traded by others.
In the Kyle model the noise traders can be considered as less than fully
rational, since they expect to suffer losses equal to the insiders’ gains. The
market makers set the prices equal to the expected value of the risky asset
conditional on the order flow; they are making zero profits. The market
makers cannot distinguish the trading of the insider from the trading of the
noise traders, who in effect provide camouflage, which enables the insider to
make profits on their expense.
The market maker in the standard model has substantial market power,
yet does not exploit this to his own advantage when setting the price; the
market maker is assumed to be a fiduciary acting in the best interest of
market participants.
One may ask how realistic this assumption is. In the testimony before
the Financial Crisis Inquiry Commission, Goldman CEO Lloyd Blankfein laid
out the Goldman Sachs perspective on the firm’s role in CDO deals related
to the 2008 financial crisis. From his answer it seems clear that he does not
consider a market maker as a fiduciary agent:
In our market-making function, we are a principal. We represent the
other side of what people want to do. We are not a fiduciary. We are not an
agent. Of course, we have an obligation to fully disclose what an instrument
is and to be honest in our dealings, but we are not managing somebody else’s
money.
Caveat emptor seems to be Mr. Blankfein’s message, and this was also the
basis of Goldman’s defense against the SEC suit re the Abacus transactions.
The case of Goldman Sachs is, we believe, not unique. Investment banks
and other financial intermediaries are known to accumulate large fortunes,
which should be difficult, or even impossible, if they were just disinterested
auctioneers.

2
 In this paper, we investigate the consequences of relaxing the assump-
tion that market makers are fiduciaries. In our model, market makers are
economic agents allowed to make a profit. Market makers generate prof-
its by adding a margin to the conditional expected value of the risky asset
when they are going short. Similarly they are subtracting a margin when
taking long positions. Formally, the margin is a random variable, which is
correlated with aggregate demand. Thus, market makers are not just adding
or subtracting a fee; the size of the fee depends on trading volume. As in
the standard model, informed traders realize what market makers are up to,
and take their behavior into account when deciding their own trades. Noise
traders just trade, we could allow them to have partial information as in
Aase, Bjuland and Øksendal (2012a), but we have chosen to let these agents
be uninformed. Despite of this, market makers may make unbounded profits
taking advantage of noise traders, which would not make sense. To avoid
this outcome a regulator is introduced. Alternatively, the market maker may
be assumed to practice restraint in order to keep markets open.
These issues were addressed in a recent paper (Aase and Gjesdal (2017)),
in the setting of a one-period model. It is of interest to extend this analysis to
several periods, which we do in this paper, where we consider a continuous-
time model.
As in the one-period model our analysis shows, perhaps surprisingly, that
for only a moderate correlation with the aggregate demand, the profits of
the market maker may exceed that of the perfectly informed insider. In the
paper we illustrate the time profiles of these profits. This could serve as one
explanation of why so much wealth tends to end up in the hands of financial
intermediaries, a timely question that has been asked many times over after
the 2008-financial crisis.
Another implication is that the market maker’s actions leads to a more
volatile price then would be the case if dealings were fair. This may throw
some light on the observation made by Campbell and Schiller (1988): Stock
market prices display much more volatility than implied by dividends alone.
The more recent approach to this problem is to consider the variations in
the stochastic discount factor, see e.g., Campbell (1999). We try to see if
the effects of the actions of market makers are substantial enough to explain
parts of this problem.
To limit the distortion of prices, a regulatory authority (the SEC) impose
an upper bound on price volatility. In our model this limits the degree to
which the market maker can perturb the price, and allows us to find an

3
equilibrium in which the insider maximizes profits and the market maker
trades ”fees”. Even if the regulatory constraint limits the market maker’s
degree of price distortion, still the market maker’s profit may exceed that of
the perfectly well informed insider. This happens for reasonable degrees of
price distortions, a consept developed in the paper.
Our pricing functional is nonlinear, which seems like a popular topic in
itself in parts of the extant literature, together with ”model uncertainty” and
similar issues. In our model the nonlinearity stems from a specific economic
assumption, namely that the market maker trades fees. As we know, in neo-
classical equilibrium theory prices are linear for a variety of reasons, among
others to avoid arbitrage possibilities, which is not an issue here.
There is a rich literature on the one period model, as well as on discrete
time insider trading, e.g., Holden and Subrahmanyam (1992), Admati and
Pfleiderer (1988), and others, all adding insights to this class of problems.
Glosten and Milgrom (1985) present a different approach, containing sim-
ilar results to Kyle. Before Kyle (1985) and Glosten and Milgrom (1985)
there is also a huge literature on insider trading in which the insider acts
competitively, e.g., Grossman and Stiglitz (1980).
The approach of this article is to study the continuous-time model di-
rectly, not as a limiting model of a sequence of auctions, and use the machin-
ery of infinitely dimensional optimization, directional derivatives (or calculus
of variations) and filtering theory to solve the problem. We also consider al-
ternatively the stochastic maximum principle in the setting of differential
game theory, as well as the Bellman approach in two appendices.
We are able to find the price of the risky asset, the various profits of the
participants, all in terms of the insider’s trading intensity. The latter we show
satisfies an integral equation, that can be solved by an iterative procedure.
This we illustrate numerically, by graphs of the the trading intensity, the
profits of the agents, and the other key variables developed in in the paper.
The paper is organized as follows: The model is explained in Section 2.
The analysis of the continuous time model starts in Section 3, where the
mean, the variance and the covariances of the order flow y is derived in Sec-
tion 3.1, with preliminary expressions for the profit functions of the insider
and the market maker. In Section 4 the insider’s optimization problem is
solved in Theorem 1, resulting in final expressions for the various profit func-
tions, as well as the other quantities of interest. In Section 5 we suggest how
the regulator’s problem may be solved, in Section 6 we introduce a measure
of price informativeness in the market, and present numerical illustrations.

4
Section 7 presents various graphs and computations, which illustrate the key
quantities in the paper. This is where we demonstrate our main conclusions.
In Section 8 we provide some suggestions for further research, and Section 9
concludes. The paper also contains four appendices.

2 The Model
At time T there is to be a public release of information that will perfectly
reveal the value of an asset; cf. fair value accounting. Trading in this asset
and a risk-free asset with interest rate zero is assumed to occur continuously
during the interval [0, T ]. The information to be revealed at time T is rep-
resented as a signal ṽ, a random variable which we interpret as the price
at which the asset will trade after the release of information. This informa-
tion is already possessed by a single insider at time zero. The unconditional
distribution of ṽ is assumed to be normal with mean µṽ and variance σṽ2 .
In addition to the insider, there are noise (liquidity) traders, and risk
neutral market makers. The noise traders are unable to correlate their orders
to the insider’s signal ṽ. Thus the noise traders have random, price-inelastic
demands. All orders are market orders and the net order flow is observed by
the market maker. We denote by zt the cumulative orders of noise traders
through time t. The process zt is assumed to be a Brownian motion with
mean zero and variance rate σt2 , i.e., dzt = σt dBt , for a standard Brownian
motion B defined on a probability space (Ω, P ). As Kyle (1985) and Back
(1992) we assume that B is independent of ṽ. We let xt be the cumulative
orders of the informed trader, and define

(2.1) yt = xt + zt for all t ∈ [0, T ]

as the total orders accumulated by time t.

The market maker only observes the process y, so he cannot distinguish
between informed and uninformed trades. Let Fty = σ(ys ; s ≤ t) be the in-
formation filtration of this process. The risk neutral market maker, assumed
to have some degree of monopoly power, sets the price pt at time t as follows

(2.2) pt = E[ṽ + ut |Fty ] := mt + E[ut |Fty ],

where mt = E[ṽ|Fty ] is the ”fair value”, and ut = kt yt for kt ≥ 0 a determinis-

tic function satisfying kt → 0 as t → T . We assume that kt = (T −t)κ, where

5
κ is a non-negative constant set by the market maker. Clearly E[ut |Fty ] =
kt yt . The market maker, the insider and the noise traders all know the prob-
ability distribution of ṽ.
We assume that the insider’s market order at time t is of the form

(2.3) dxt = (ṽ − pt )βt dt, x0 = 0,

where β ≥ 0 is some deterministic function. This form of the market or-

der follows from the discrete time formulation of the problem, assuming the
insider maximizes profits, in which case (2.3) follows from the first order con-
dition; xt does not depend on pt since xt is submitted before pt is set by the
market makers.
Assumption (2.3) is consistent with Kyle (1985).1 The function βt is called
the trading intensity on the information advantage (v − pt ) of the insider.
The basic assumptions behind this result is (i) profit maximization by
the insider, where it is shown in Aase and Gjesdal (2016) that this result
still holds when the market maker sets the price as we have assumed in
(2.2) above, and (ii) the insider does not condition the quantity he trades
on price. Here the insider chooses quantities (”market orders”) instead of
demand functions (”limit orders”).
Assumption (2.2) is our deviation from the standard model.2 Below we
explain why this price setting leads to a positive expected profit for the
market maker.
The stochastic differential equation for the total order yt is

(2.4) dyt = (ṽ − mt )βt dt − kt βt yt dt + σt dBt .

Aside from the first mean zero ’innovation’ term, the equation shows that
yt has the structure of a (generalized) mean reverting Ornstein-Uhlenbeck
process, oscillating around this mean zero term.
Let us denote by St (β) = E{(ṽ −pt )2 } and by γt (β) = E{(ṽ −mt )2 }. Usu-
ally the assumption is made that limt→T − pt = pT = ṽ a.s. This assumption
seems natural, ensuring that all information available has been incorporated
in the price at the time T of the public release of the information, at which
time a price spread cannot be sustained.
1
The finite variation property of x is assumed by Kyle (1985), and an equilibrium where
this is the case is found by Back (1992).
2
An alternative would be to assume that the market maker is risk averse, which would
lead to a different model.

6
 In Aase et. al. (2012a) mt = pt for all t ∈ [0, T ], and it was there
demonstrated that pt → ṽ as t → T − , and St (β) → 0 as t → T − as a
consequence of the insider following his optimal trading strategy. Here we
find it natural to simply assume this, as was done in e.g., Kyle (1985), so that
pt − mt → 0 as t → T − , and both converge to ṽ, since kt → 0 by assumption.
Denote the insider’s wealth by w and the investment in the risk-free asset
by b. The budget constraint of the insider can best be understood by con-
sidering a discrete time setting, of which the continuous-time model is the
limit (in an appropriate sense). At time t the agent submits a market order
xt −xt−1 and the price changes from pt−1 to pt . The order is executed at price
pt , in other words, xt −xt−1 is submitted before pt is set by the market makers.
The investment in the risk-free asset changes by bt −bt−1 = −pt (xt −xt−1 ), i.e.,
buying stocks leads to reduced cash with exactly the same amount. Thus,
the associated change in wealth is

(2.5) bt − bt−1 + xt pt − xt−1 pt−1 = xt−1 (pt − pt−1 ).

In other words, the accounting identity for the wealth dynamics is of the
same type as in the standard price-taking model, except for one important
difference; while, in the rational expectations model, the number of stocks
in the risky asset at time t depends only on the information available at this
time, so that both the processes x and p are adapted processes with respect
to the same filtration, here the order x depends on information available only
at time T for the market makers (and the noise traders).
As a consequence of (2.5) we obtain the dynamic equation for the insider’
wealth wtI as follows
Z t
I I
(2.6) wt = w0 + xs dps
0

This is not well-defined as a stochastic integral in the traditional inter-

pretation, since pt is Fty -adapted, and xt is not. Thus it needs further ex-
planation. However, since we assume that the strategy of the insider has the
form (2.3) for some deterministic continuous function βt > 0, then a natural
interpretation of (2.6) is obtained by using integration by parts, as follows:

7
 Z t
wtI = w0I
+ xt p t − ps dxs
0
Z t Z t
I
= w0 + pt (ṽ − ps )βs ds − ps (ṽ − ps )βs ds
0 0
Z t Z t
I 2
(2.7) = w0 + (ṽ − ps ) βs ds − (ṽ − pt )(ṽ − ps )βs ds.
0 0

Alternatively, one might obtain (2.7) by interpreting the stochastic inte-

gral in (2.6) as a forward integral. See Russo and Vallois (1993), Russo and
Vallois (1995, 2000) for definitions and properties and Biagini and Øksendal
(2005) for applications of forward integrals to finance.
Similarly we can find the market maker’s profit from his price setting
operations: His wealth wM from these operations is

wM = w0M + (p0 − p1 )y0 + (p1 − p2 )y1 + · · ·

When the total initial order y0 > 0, the market maker has to sell to clear
the market and accordingly sets the price p0 a bit higher than he would have
done if he were a fiduciary. Similarly, if y0 < 0 she must buy to clear the
market, so he sets the price p0 a bit lower than he would if he sets the price
fairly. Continuing this practice in every period, he will end up with a positive
expected profit, simply because the profit he would have obtained by being
fair has zero expectation3 .
Consider the situation where the total initial order y0 > 0. Because of the
mean reverting nature of y towards zero, it is more likely that y1 < y0 than
the other way around. By the price setting mechanism used by the market
maker, it is more likely that p1 < p0 than the opposite, in which case the
market maker’s profit is positive. A similar reasoning holds when y0 < 0, in
which case the market maker buys from the other participants at time zero,
and sells the stock in the market at time one at the price p1 he sets then,
based on y1 − y0 . Thus, in expectation the market maker’s profit is positive.
Notice that the market maker takes some ’overnight’ risk, in that, when
he must sell to the other participants at time t, he sets the price pt which he
sells for, and the next day he sets the price pt+1 , based on the order yt+1 − yt ,
3
One may think of trade as ”synthetic” in that only money changes hands, based on
dynamics of the underlying stock.

8
at which he buys in the market the stock that he ’delivered’ the day before.
By the price setting mechanism, more likely than the other way he profits
from this operation. If he were a fiduciary, he would go even in ’the long
run’. Here as a non-fiduciary, in expectation his profit is positive.
By going to the continuous time limit, his wealth at time t is
Z t Z t
M M M
(2.8) wt = w0 − ys dps = w0 − pt yt + ps dys + [p, y]t ,
0 0

where [p, y]t is the quadratic covariance process of p and y. Unlike the cor-
responding expression for the insider, this integral is well-defined in the tra-
ditional interpretation, since pt is Fty -adapted, and so is of course yt .
Finally, the noise traders’ profit is
Z t Z t
N N N
(2.9) wt = w0 + zs dps = w0 + zt pt − ps dzz − [p, z]t .
0 0
Rt
The stochastic integral 0 zs dps is well-defined in the traditional meaning
since zt is FtB -adapted, pt is Fty -adapted and Fty ⊃ FtB , and hence, by
integration by parts, so is the latter stochastic integral in (2.9).
Since yt = xt + zt and x is of bounded variation, [p, y]t = [p, z]t for all t.
Since this is a pure exchange economy, it follows that the sum of the profits
is zero with probability one, or, wtI + wtM + wtN = w0I + w0M + w0N a.s.

3 Some basic analysis.

Returning to the stochastic process for the total order at time t, yt , its
representation is given by (2.4), which we repeat here


dyt = ṽ − E(ṽ|Ft ) βt dt − kt βt yt dt + σt dBt .

This is a Gaussian process consisting of an Ornstein-Uhlenbeck type process,

with a normally distributed ”innovation” term added to its drift term, the
first term on the right-hand side in the above stochastic differential equation.
In order to analyse this model for the total order, we start by rewriting
this equation as follows:

dyt + yt kt dt = (ṽ − mt )βt dt + σt dBt .

9
If we define
Z t
(3.1) ỹt := yt exp( ks βs ds)
0

and use Ito’s lemma, we obtain the following

Z t Z t
(3.2) dỹt = (ṽ − mt )βt exp( ks βs ds)dt + σt exp( ks βs ds)dBt .
0 0

Clearly F (y) = F (ỹ) and hence

(3.3) mt = E[ṽ|F (y) ] = E[ṽ|F (ỹ) ].

Therefore we may regard (3.2) as the innovation process of an ”observation

process” ŷt defined by
Z t Z t
(3.4) dŷt = ṽβt exp( ks βs ds)dt + σt exp( ks βs ds)dBt ; ŷ0 = 0.
0 0

For this to hold, we need to verify that

(ỹ) (ŷ)
(3.5) Ft = Ft .

Suppose (3.5) is proved. Then

(ŷ)
mt = E[ṽ|Ft ]

is the filtered estimate of v given the observations ỹs ; s ≤ t.

By Theorem 12.1 in [18] or Theorem 6.2.8 in [19], the filter mt is given
by the SDE
Rt Z t
γt βt exp( 0 ks βs ds) h i
(3.6) dmt = 2 Rt dŷt − βt exp( ks βs ds)mt dt ; t ≥ 0;
σt exp(2 0 ks βs ds) 0

m0 = E[ṽ],
(β)
where St = St = γt (β) + kt2 V (t), where V (t) = E(yt2 ), and γt (β) solves the
Riccatti equation
βt2 γt2
(3.7) dγt = − ; t≥0
σt2
γ0 = E[(ṽ − E[ṽ])2 ].

10
Thus we have a controlled state process

(ŷt , mt , γt )

given by the equations (3.4),(3.6) and (3.7).

Rewriting the system in terms of (yt , mt , γt ) we obtain the following set
of equations

(3.8) dyt = (ṽ − mt − kt yt )βt dt + σt dBt ; y0 = 0

γt βt
(3.9) dmt = 2 [(ṽ − mt )βt dt + σt dBt ]; m0 = p0 = E[ṽ]
σt
β 2γ 2
(3.10) dγt = − t 2 t ; γ0 = E[(ṽ − E[ṽ])2 ].
σt
The expected profits are
Z T Z T
M
(3.11) J (k, β) := w0M + E( kt yt (kt yt + mt − ṽ)βt dt − yt2 dkt )
0 0
Z T
(3.12) J I (k, β) := w0I + E[(ṽ − ms − ks ys )2 ]βs ds.
0

for the market maker and for the insider, respectively.

Let us now return to the problems of the previous section and calculate
the profits of various participants in this economy. Towards this end we first
need expressions for the mean, the variance and the covariances of the market
order process y.

3.1 The variance and covariances of the process y.

We start with the variance. Based on the expression in (3.3), we proceed as
follows. From equation (3.1) we have by Itô’s lemma
1
d(ỹt )2 = 2ỹt dỹt + 2(dỹt )2 =
2
Z t Z t
2



2yt (ṽ − mt )βt exp ks βs ds dt + σt exp ks βs ds dBt +
0 0
Z t
σt2 exp 2


ks βs dt.
0

11
From this we deduce that
 2 hZ t h Z s
2


E ỹt = E 2ys (ṽ − ms )βs exp ku βu du ds
0 0
Z s Z s
i
ku βu du dBs + σs2 exp 2


+σs exp ku βu du ds =
0 0
Z t Z s
σs2 exp 2 ku βu du)ds.
0 0
Observe that

E (ṽ − mt )yt2 = E(ṽyt2 ) − E E(ṽ|Fty )yt2 = 0

since E E(ṽ|Fty )yt2 = E E(yt2 ṽ|Fty ) = E(ṽyt2 ). Hence

Z t Z t Z s
ku βu du E yt2 = σs2

 

exp 2 exp(2 ku βu du ds
0 0 0

or
Rt
Z t Ru
E yt2 = e−2 0 kr βr dr σu2 e2 kr βr dr
 
(3.13) 0 du.
0

This expression will be useful below. We use the notation V (t) := E(yt2 ) for
all t ∈ [0, T ]. 4
Moving to the covariances E(yt ys ) for any s > t, we proceed as follows.
Here we use the notation Rt
e(t) := e 0 kr βr dr .
For s > t,
ds (ỹs ỹt ) = (ṽ − ms )βs e(s)ỹt ds + σs e(s)ỹt dBs .
Integrating this from t to s, we get
Z s Z s
E[(ỹs − ỹt )ỹt ] = E[ (ṽ − mr )βr e(r)ỹt dr + σr e(r)ỹt dBr ] =
t t
Z s
E[(ṽ − mr )ỹt ]βr e(r)dr + 0 = 0,
t
4
The theory leading to the result in (3.13) may be linked to a deeper result in filtering
theory. For details, see Appendix 4.

12
since
E[(ṽ − mr )ỹt ] = E[E[(ṽ − mr )ỹt |Fty ] = E[ỹt E[ṽ − mr |Fty ]] = 0.
The latter equality follows from E[(ṽ −mr )|Fty ] = E[E[(ṽ −mr )|Fry |Fty ]] = 0,
since the inner conditional expectation is zero. We obtain for s > t
E[ỹs ỹt ] = E[(ỹs − ỹt )ỹt ] + E[ỹt2 ] = E[ỹt2 ].
Using the definition of ỹ, we have that
Rs
E[ys yt ] = E[yt2 ]e− t kr βr dr

Combining this with our above result (3.13), we conclude that

Z t
−( 0s kr βr dr+ 0t krβr dr)
R R Ru
(3.14) E[ys yt ] = e σu2 e2 0 kr βr dr du, for s > t.
0

For s = t we obtain the result in equation (3.13).

Figure 1 illustrates a graph of the covariance function C(s, t; κ) := E[ys yt ]
when κ = 0.045 for s, t ∈ [0, 10].

Fig. 1: The covariance function C(s, t) of y when κ = 0.045.

The base case parameters are σt = σ = 0.20, a constant for all t ∈ [0, T ].
Also γ0 = σṽ2 , where σṽ = 0.30, and we have chosen T = 10. (Here we have
anticipated a bit, and used the optimal value of the trading intensity βt of
the insider appearing in Section 4 below.)

3.1.1 The mean of y

We will also need the mean E(yt ) of the process y for any t. Starting with
the equation
Z t Z t Z t
yt = y0 + (ṽ − E(ṽ|Fs )βs ds − ks βs ys ds + σs dBs ,
0 0 0

13
and letting E(yt ) := ȳt , where ȳ0 = y0 , by taking expectation in the above
equation we obtain Z t
ȳt = y0 − ks βs ȳs ds
0
or
d
ȳt = −βt kt ȳt
dt
which is an ordinary, linear differential equation in ȳt , with initial condition
ȳ0 = y0 , the unique solution of which is
Rt
E(yt ) = y0 e− 0 ks βs ds
.
In our model y0 = 0, which implies that E(yt ) = 0 for all t ∈ [0, T ]. Thus
E(pt ) = E(mt ) + kt E(yt ) = E(mt ) = E(ṽ) = µṽ , so the price pt has the
correct expectation at all times.

3.2 The profit of the insider

Returning to the insider, from equation (2.7) giving the wealth wt of the
insider at any time t, since
Z T
E[(ṽ − pT )(ṽ − pt )]βt dt = 0
0

by our assumption that pt → pT = ṽ, his task is to find the trading intensity
βt which maximizes the expected terminal wealth
Z T
I I
(3.15) E[wT ] = w0 + E[(ṽ − mt − kt yt )2 ]βt dt := J I (k, β).
0

Later, when we consider the net profit at any time t ∈ [0, T ], we will use the
notation pI (t, κ) for the insider’s net profit by time t, so that J I (k, β)−w0I :=
pI (T, κ)) with this notation. Similarly for the market maker.
The dilemma for the insider is that an increased trading intensity at some
time t will reveal more information about the value of ṽ to the market makers
and hence induce a price pt closer to ṽ, which in turn implies a reduced insider
information advantage. On the other hand she has to trade in order to make
any profit at all.
First observe that
E (ṽ − mt )yt = E(ṽyt ) − E E(ṽ|Fty )yt = 0

14
since E E(ṽ|Fty )yt = E E(yt ṽ|Fty ) = E(ṽyt ), a result similar to the one

obtained above, with yt instead of yt2 .

By the definition of γt (β) = E(ṽ − mt )2 , we then obtain the following
Z T
I
(3.16) J (k, β) = w0I + βt (γt (β) + kt2 Vt )dt
0

since the cross term vanishes by by the above observation. Using the expres-
sion for V (t) := E(yt2 ) given in (3.13), we obtain the following
Z T Rt
Z t Rs
I
w0I kt2 e−2 ks βs ds
σs2 e2 kr βr dr


(3.17) J (k, β) = + βt γt (β) + 0 0 ds dt.
0 0

The insider will now maximize this expression in the trading intensity
process β, for a given price perturbation process k by the market maker.
Before we address this problem, we want to find the profit of the in-
sider at any time t ∈ [0, T ], which will allow us to observe the relative time
performance of the two profit functions of interest.
Towards this end, let us go back to the expression for the insider’s profit
at time t given in (2.7). Taking expectation in this equation we obtain
Z t Z t
I I 2
E(wt ) = w0 + E(ṽ − ps ) βs ds − E(ṽ − pt )(ṽ − ps ) βs ds =
0 0
Z t Z t
w0I + (γs (β) + kt2 V (s)) βs ds − E(ṽ − mt − kt yt )(ṽ − ms − ks ys ) βs ds,
0 0

where the second term follows from (3.17). Consider the last term. The
integrand can be written

(3.18) E(ṽ − mt − kt yt )(ṽ − ms − ks ys ) = E(ṽ − mt )((ṽ − ms )−

ks E((ṽ − mt )ys ) − kt E((ṽ − ms )yt ) + kt ks E(yt ys ).

The second expectation on the right-hand side is

E((ṽ − mt )ys ) = E[E[(ṽ − mt )ys |Fsy ]] =

E[ys E(ṽ − mt |Fsy )] = E[ys E[E(ṽ − mt |Fty )]|Fsy ] = 0

by standard iterated expectations, since E(ṽ − mt |Fty ) = 0, as shown before.

15
 Notice that s ≤ t in these computations. The third expectation on the
right-hand side of (3.18) is

E((ṽ − ms )yt ) = E[E[(ṽ − ms )yt |Fsy ]]

= E[E[E(yt (ṽ − mt )|Fsy ]|Fty ] = E[yt E[E(ṽ − ms |Fsy )]|Fty ] = 0,

where the second equality above follows from a not so standard, but rather
obvious, iterated expectation result (see e.g., Tucker (1967), Th 6, Ch 7),
and again, because E(ṽ − ms |Fsy ) = 0, the result follows.
It remains to compute the first expectation on the right-hand side of
(3.18). It follows from Theorem 3.1 in Aase and Øksendal (2018) that

E(ṽ − mt )((ṽ − ms ) = γt (β).

The last term in (3.18), the covariance, we have already computed in Section
3.1. Since here t ≥ s, we rewrite this formula accordingly, namely as
Z s
−( 0t kr βr dr+ 0s krβr dr)
R R Ru
(3.19) E[yt ys ] = e σu2 e2 0 kr βr dr du, for t ≥ s.
0

This means that the insider’s profit at any time t in [0, T ] is given by
Z t Z t Z t
I I 2
E(wt ) = w0 + (γs (β) + ks Vs )βs ds − γt (β) βs ds − kt ks E(yt ys )ds.
0 0 0

Observe that as t → T , this profit converges to the expression in (3.16), since

both γt (β) → 0 and kt → 0 then. By use of (3.19) the insider’s profit can be
written
Z t Z t
I I 2
(3.20) E(wt ) = w0 + (γs (β) + ks Vs )βs ds − γt (β) βs ds
0 0
Z t Z s
− 0t kr βr dr − 0s kr βr dr
R R Ru
σu2 e2 0 kr βr dr du ks ds.


− kt e e
0 0

The problem of finding the optimal value of the insider’s trading intensity
βt , and the corresponding expression for the profit fundtion is relegated to
Section 4 below.

16
3.3 The profit of the market maker
The market maker’s expected profit is:
Z T
M
E(wTM ) w0M


J (k, β) := = −E yt dpt =
0
Z T Z T
w0M yt2 dkt =


−E kt yt dyt +
0 0
Z T Z T
w0M yt2 dkt =


−E kt yt (ṽ − mt − kt yt )βt dt +
0 0
Z T Z T
w0M + kt2 (Eyt2 )βt dt + κ Eyt2 dt.
0 0
The third equality follows since m is a martingale, the fourth since Bt is a
Fty -martingale, and the last equality follows since yt is orthogonal to (ṽ −mt ),
and the Fubini theorem. Thus we have that this profit can be written
Z T
M M
ks2 V (s)βs + κV (s) ds.


(3.21) J (k, β) = w0 +
0

Notice that the profit of the market maker at any time t ∈ [0, T ] is simply
Z t
M M
ks2 V (s)βs + κV (s) ds.


(3.22) E(wt ) = w0 +
0

Using the expression (3.13) for Vs = E(ys2 ), we obtain the following ex-
pression for this profit:
Z T  Rs
Z s Ru
M
(3.23) J (k, β) = w0M + (ks2 e−2 0 kr βr dr
σu2 e2 0 kr βr dr
du)βs
0 0
Rs
Z s Ru 
+ κ(e−2 0 ke βr dr
σu2 e2 0 kr βr dr
du) ds.
0

Consider the latter profit. The last term on the right-hand side increases
without bounds as kt = (T − t)κ increases without bound for any given t,
i.e., as the constant κ → ∞. Surely kt goes to zero as t goes to T , but the
constant κ can in principle be set arbitrarily large by the market maker, since
she simply decides the value of this constant once and for all. Also we know

17
that βt decreases with κ, but this effect more or less cancels out since the
two exponentials where β enters are of different signs. R
T
Likewise, the second term on the right-hand side, 0 kt2 (Eyt2 )βt dt, also
possesses this property, despite the fact that here β enters linearly (in addi-
tion to its exponential dependence).
This is illustrated numerically in Figure 2. The base case parameters
are the same as in Figure 1, where the horizon is T = 10. (Again we have
anticipated a bit, and used the optimal values of the function βt appearing
is Section 4 below.)
Using the notation for the net profit of the market maker
Z t Z s
2 −2 0s kr βr dr
R Ru
pM (t, κ) := (ks e σu2 e2 0 kr βr dr du)βs
0 0
Z s 
−2 0s ke βr dr
R Ru
+ κ(e σu2 e2 0 kr βr dr du) ds.
0
at the intermediate time t ≤ T , the upper graph is the net, terminal profit
pM (T, κ) as a function of κ, while the the lower graph shows the net profit
pM (t, κ) accumulated at the intermediate time t = 2 as a function of κ.

Fig. 2: The profit functions of the market maker as a function of κ.

As a result, this model displays similar properties to the one-period model,
and a regulator is therefore introduced to limit the price perturbation caused
by the market maker trading fees.
This set-up does not become a game between the insider and the market
maker in the usual meaning of game theory, in that only the insider maxi-
mizes an objective, while the market maker trades fees that depend on the
stochastic order flow, i.e., she sets the price to the best of her knowledge, and
then adds the fee conditional on observing the order flow. In some sense, the
market maker is not strategic in the ordinary interpretation of this term.

18
 As the market maker obtains more information from the order flow, she
lets this information be reflected in the price pt . The introduction of trading
fees reduces the informational contents of the true value of the asset in the
price. The market maker may be assumed to set κ to the maximum value
allowed by the regulator, or alternatively, by her own conscience, supposing
she practices restraint in order to keep the markets open, whichever gives
the smallest value of κ. It is in the interests of the market maker that the
market does not break down, in which case she does not make any profits at
all, and may also face legal issues. It is, after all, the market maker’s task to
operate such that the markets function.
The problem of relating the parameter κ to observables in the market is
treated in Section 5 below.
Since this is a pure exchange economy, the profit of the noise traders is
given by
J N (k, β) = w0I + w0M + w0N − J I (k, β) − J M (k, β).
They will loose in this market.

4 The insider’s problem

We now address the optimization problem of the insider. In our framework
he is to determine the trading intensity βt by which he trades at each time
t ∈ [0, T ]. We assume he determines this intensity such that his profit J I (k, β)
is maximized, taking k as given. Vi have that

dpt = dmt + d(kt yt ) = dmt − κyt dt + kt dyt ,

since the function kt is of bounded variation. From filtering theory (see e.g.,
Kalman (1960), Davis (1977-84), Kallianpur (1980) or Øksendal (2003), Ch
6) we know that the corresponding conditional expected value mt = E(ṽ|Fty )
is given by
βt γt (β)
dmt = dyt .
σt2
Furthermore the square error function γt (β) = E(ṽ−mt )2 satisfies the Ricatti
equation
d β2
γt (β) = − t2 γt2 (β),
dt σt

19
which has the solution
σṽ2
(4.1) γt (β) = 2 t
R ,
1 + σṽ 0
β̃s2 ds
where β̃t := σβtt . Here γ0 = E(ṽ − Eṽ)2 = σṽ2 . Accordingly, the insider’s
problem is to solve the following
Z T Z t
σṽ2 βt 2 −2 0t ks βs ds
R
2 2 0s kr βr dr
R 
(4.2) supβ t 2
+ βt k t e σ s e ds dt.
1 + σṽ2 0 βσs2 ds
R
0 0
s

We find it natural to use directional derivatives, or equivalently, variational

calculus to solve this problem. Based on this we have the following:
Theorem 1. The optimal trading intensity βt of the insider satisfies the
following integral equation
Z T
σt2  Rs

(4.3) βt = R T 2
γt (β) − V (t) kt + 2kt βs ks2 e−2 t kr βr dr ds ,
2 t γs (β)2 βs ds t

where V (t) is the variance process of the order flow yt .

Proof: The proof can be found in Appendix 1.5
This integral equation can be solved iteratively, which we demonstrate in
Section 7 below.
When κ = 0, the trading intensity is seen to be
σt2 γt (β)
(4.4) βt0 = RT , (κ = 0.)
2 t
γs (β)2 βs ds
This can further be reduced to the following simple expression (see Aase et.al
(2012a,b))
R  21
T
σt2 0 σs2 ds
(4.5) βt0 = RT , when κ = 0.
σṽ t σs2 ds
5
The problem may alternatively be formulated in terms of a stochastic differential
game between the insider and the market market maker, in which case we make use
of the stochastic maximum principle. This leaves three adjoint variables (co-variables
with shadow price interpretations) to be determined. Alternatively we can formulate the
problem as a dynamic programming problem and use the Bellman approach. In this case
this leaves us with the indirect profit function to be determined. We indicate these two
formulations later (Appendix 2 and 3), without going all the way to the the bitter end.

20
When σt = σ for all t ∈ [0, T ], where σ is a positive constant, this finally
reduces to the Kyle (1985)-solution.

5 The regulator’s problem

To limit the distortion of prices, a regulatory authority (the SEC) imposes
an upper bound on price volatility. This is by and large the same as limiting
the conditional expected degree of price distortion (see Aase and Gjesdal
(2018)). In our model this limits the market maker’s freedom to set prices.
The market maker in our model is not really strategic, is risk neutral but
exercises a certain degree of monopoly, as explained earlier. The regulator is
introduced to mitigate this.
As in the standard model, informed traders realize what the market maker
is up to, and take his behavior into account when deciding their own trade.
Noise traders just trade. In this situation the market maker can make un-
bounded profits taking advantage of noise traders, which would not make
sense. To avoid this outcome the regulator is introduced.
It is well acknowledged that insider trading increases the volatility of an
asset. Also price perturbations caused by the market maker’s trading fees
increase the volatility. This can be utilized by the regulator, who can suspend
the stock from further trading based on observing volatility over a certain
acceptable, preset limit. A measure of volatility we consider as the basis for
the regulator’s ability to monitor the market.
The decision variable κ of the market maker has so far ”no dimension”,
meaning that it is not an observable quantity in the market. We therefore
seek a connection between this variable and and an observable quantity. This
is an important step in the analysis, because it allows us to see if the market
maker can really outperform the perfectly well informed insider in terms of
profits at reasonable levels of trading fees, i.e., at levels where the regulator
has not suspended the security.
From our expressions for the profit functions of the participants, we notice
that as κ increases, the market maker’s profit grows without limits, see e.g.,
Figure 2, and eventually it will dominate the profit function of the insider.
The interesting question is then if this takes place at an acceptable level of
price perturbations, set by the regulator.
With this in mind, we would like to develop a connection between the
decision variable κ and total volatility. Consider the quantity var(pt ) =

21
E(pt − E(ṽ))2 . It is closely connected to the mean square deviation

St (β) = E(pt − ṽ)2 = γt (β) + kt2 E(yt2 ).

This latter quantity, or its square root, we assume can be observed by the
regulator, who will then compare this to the corresponding term γt (β 0 ) based
on no price distortions by the market maker.
Recall the following definitions. St (β) = E(ṽ − pt )2 and γt (β 0 ) = E(ṽ −
mt )2 where mt = E(ṽ|Fty ). The function γt (β 0 ) corresponds to the expected
square deviation between the true value of the asset and the fiducial price
mt , provided the trading intensity βt0 is used in the computation of the latter
quantity. St is the expected square deviation between the true value of the
asset and the actual price that the market maker sets, in the case where
she trades fees, as explained. Naturally, St (β) is larger than γt (β 0 ), and
increasingly so as the market maker’s decision variable κ increases.
This leads us to introduce the following quantity in relative terms (rv =
relative volatility)
p
St (β)
(5.1) rv(t, κ) := p , t ∈ [0, T ], κ ≥ 0.
γt (β 0 )

Our assumption that St := St (β) is observable by the regulator also means

that rv(t, κ) is observable.
From our previous results St (β) = γt (β) + kt2 Vt (k), where V (t) depends
on kt and is given by equation (3.13), which is
Z t
−2 0t ks βs ds
R Rs
2
V (t) = E(yt ) = e σs2 e2 0 kr βr dr ds,
0

and from (4.1) we have that

σṽ2
γt (β 0 ) = R t ,
1 + σṽ2 0 (β̃s0 )2 ds
β0
where β̃t0 := σtt , and γ0 = E(ṽ − Eṽ)2 = σṽ2 . Using these relations, rv(t, κ)
can be written
 21
kt2 −2 R t ks βs ds t 2 2 R s kr βr dr
 Z
γt (β)
(5.2) rv(t, κ) = + e 0 σs e 0 ds .
γt (β 0 ) γt (β 0 ) 0

22
When κ = 0 we see from this expression that rv(t, 0) ≡ 1.
Here rv will give information about the degree to which the market maker
trades fees. When κ = 0, the function rv(t, 0) is constant through time and
identically equal to 1 as noticed. As κ increases from zero, rv(t, κ) will
rise above 1, and indicate the percent-wise increase from the situation with
fiducial trade, at every t ∈ [0, T ].
For example, when rv(t, κ) = 1.20, for some t and κ, the actions of the
market maker has increased the volatility of the asset by 20% relative to
the situation with fiducial price setting, where κ = 0. Thus the quantity rv
seems like a reasonable measure of the degree of fee trading, in the hands of
a regulator.
In this situation the map from κ to rv, and in particular its inverse
mapping, will serve as a guide for acceptable values of κ, given a certain
level of rv set by the regulator. This inverse mapping is illustrated in Section
7.5 below6 . The market maker can, on the other hand, use this mapping to
exercise restraint in setting κ in order to keep markets open.

6 A measure of price informativeness

We now derive a measure of the informativeness in prices, which is of partic-
ular interest when the prices are distorted. Consider the quantity

var(ṽ|pt )
ι(t, κ) := 1 − .
var(ṽ)

When the price carries no private information about the true value of the
asset at some time t and some level of distortion κ, the conditional variance
equals the unconditional variance, and consequently ι(t, κ) = 0 at this point
(t, κ). When the price equals the value of the asset, the conditional variance
equals 0 and ι = 1 at this point, in which case all the private information
is reflected in the price. Consequently 0 ≤ ι(t, κ) ≤ 1 for all time points
t ∈ [0, T ] and for all κ ≥ 0.
Because of the joint normal assumption,

var(ṽ|pt ) = var(ṽ)(1 − ρ2ṽ,pt )

6
In this section numerical illustrations can also be found, see in particular the two first
rows in Table 2.

23
where ρṽ,pt is the correlation coefficient between ṽ and pt . Consequently
ι(t, κ) = ρ2ṽ,pt for all (t, κ).
In order to find this measure of price informativeness, we need to com-
pute the quantities cov(ṽ, pt ) and var(pt ). To this end, we first consider the
covariance. Since pt = mt + kt yt ,

cov(ṽ, pt ) = cov(ṽ, mt ) + kt cov(ṽ, yt ).

The first term on the right-hand side can be written

cov(ṽ, mt ) = E(ṽmt ) − E(ṽ)E(mt ) = E(E(ṽmt |Fty )) − µ2ṽ =

= E(mt E(ṽ|Fty )) − µ2ṽ = E(m2t ) − (E(mt ))2 = var(mt ),

since mt is Fty -measurable, where mt = E(ṽ|Fty ) and E(mt ) = µṽ . Further-
more,

γt (β) = E(ṽ − mt )2 = E(ṽ 2 ) − E(m2t ) = var(ṽ) − var(mt ),

by a similar type of conditioning as above. The last equality follows since

E(ṽ) = E(mt ) = µṽ . Since we already have an expression for γt = γt (β), see
equation (4.1), we now have an expression for var(mt ), and hence cov(ṽ, mt ) =
var(mt ) = σṽ2 − γt (β) by the above.
The term cov(ṽ, yt ) is calculated as follows: First notice that by iterated
expectations cov(ṽ, yt ) = cov(mt , yt ). From the Kalman filter equation we
have
βt γt (β
dmt = dyt ,
σt2
see equation (3.9), and from the binormality between mt and yt and the
corresponding projection theorem we obtain the following connection

βt γt (β)σyt
ρmt ,yt = ,
σt2 σmt
p
were ρmt ,yt p
is the correlation
p coefficient between mt and yt , σyt := V (t)
and σmt := var(mt ) = σṽ2 − γt (β).
We have then shown that
q p
2
cov(ṽ, pt ) = σṽ − γt (β) + kt ρmt ,yt σṽ2 − γt (β) V (t),

24
where we have formulas for all the terms on the right-hand side of this equa-
tion.
It remains to find the variance of pt . Again, from pt = mt + kt yt it follows
that

(6.1) var(pt ) = var(mt ) + kt2 var(yt ) + 2kt cov(mt , yt ) =

q p
σṽ2 − γt (β) + kt2 V (t) + 2kt ρmt ,yt σṽ2 − γt (β) V (t).

Putting all this together, we have

p p
σṽ2 − γt (β) + kt ρmt ,yt σṽ2 − γt (β) V (t)
(6.2) ρṽ,pt = p
σṽ var(pt )
where var(pt ) is given above in (6.1). From this the informativeness ι in
prices is given by
(6.3) ι(t, κ) = (ρṽ,pt )2 ,
where ρṽ,pt is calculated using (6.2) and (6.1).
Table 1 illustrates the time development of the informativeness in the
market. For a given value of the price distortion parameter κ we notice that
the informativeness increases with time, see the last row in Table 1.
In the same table we have also computed some of the other key quanti-
ties that is used in the computation of the measure of informativeness ι(t, κ),
such as the correlation coefficients ρmt ,yt (t, κ) and ρṽ,pt (t, κ) and the variances
of the price pt and the ’fair’ price mt . While ι(t, κ) = ρṽ,pt (t, κ)2 and thus
ρṽ,pt (t, κ) represents an equivalent measure of information as ι(t, κ), the cor-
relation coefficient ρmt ,yt (t, κ) is a measure connected to the ’fair’ value mt
instead of the market price pt , but computed with the value of βt where the
insider has optimally adjusted to the actual distortion of the price.
This measure tells us how closely correlated the ’fair’ price mt is to the
order flow yt . From the table we notice that there can be a high correlation,
as in our example, which throws some new light on the market maker’s
advantage in observing the order flow.
In the example this measure decreases with time up to a certain value
∗
t , then increases after that, so that here the advantage is highest in the
beginning and towards the end of the trading interval.
Since the correlation cov(ṽ, mt ) = var(mt ), this covariance also increases
slowly with time in our example, which is reasonable since more information

25
about the true value of the asset becomes available with p time, hence the
increase in the corresponding correlation coefficient ρṽ,mt = cov(ṽ, mt )/σṽ .
At the same time this shows that γt (β) decreases slowly with time, again for
the same reason: One knows more about the true value as time increases, and
the present measure of uncertainty (the mean square error) then naturally
decreases.
Finally, we illustrate the time development of the variance of the price pt
and of the ’fair’ price mt . The latter one has just been explained. For the
former two different effects are in force: One is that V (t) increases with time
(see Section 7.3 below) and what has just been shown, that cov(ṽ, mt ) also
increases with time. The other is that kt decreases with time, see the expres-
sion (6.1) for the var(pt ). In our example the first effect weakly dominates.

Cont. model:
t 1 2 3 4 5 6 7 8 9 10
ρmt ,yt (t, κ) .96 .92 .87 .84 .81 .79 .78 .79 .83 1.00
ρṽ,pt (t, κ) .31 .44 .53 .61 .67 .73 .79 .84 .91 .99
var(pt ) .02 .04 .05 .06 .07 .07 .08 .08 .08 .09
var(mt ) .009 .02 .03 .03 .04 .05 .06 .06 .07 .09
ι(t, κ) .10 .19 .28 .37 .45 .54 .62 .71 .83 1.00

Table 1: The quantity ι(t, κ) as a function of time (κ = 0.035).

The shape of ι(t, κ) as a function of the price distortion parameter κ for

a given value of time t is illustrated in the next section, see the last row
of Table 2 below. Naturally we expect that ι(t, κ) decreases with κ for any
given value of t.

7 Illustrations of the theoretical results

7.1 General
Based on the integral equation (4.3) for trading intensity of the insider, we
now present a few illustrations.
First, we indicate how to deal with this integral equation. This equation
can be transformed to a differential-integral equation if one so pleases, but we
choose to work with the version we have, where we use an iterative procedure.

26
 As a first step we suggest to use a trial solution, βt0 say, on the right-hand
side of this equation, and then find the first approximation, βt1 , given by the
left hand side as a function of this initial solution. A reasonable candidate
for the trial solution is of course the solution when κ = 0, which we have
in closed form (see (4.5)). Next one continues this procedure, where β1 (t)
becomes the new trial solution in the next step, and so on until convergence.
This, of course, requires the right numerical tools.

7.2 The trading intensity of the insider

In Figure 3 below we illustrate the time development of the trading intensi-
ties of the insider obtained this way, where the upper curve is when κ = 0
and the lower curve is for κ = 0.045. The other base case parameters are the
same as in Figure 1. We choose T = 10 here as well7 .

Fig. 3: The insider’s trading intensities as functions of t.

Note that in both cases the insider intensifies her trade towards the hori-
zon. Also, it is reasonable that the lower graph corresponds to the insider’s
trading intensity when the market maker perturbs the price. The insider,
knowing what the market maker is up to, now trades more softly. However,
towards the end she picks up trading again, since then the market maker
trades fees to a less and less extent as t approaches T .
This type of analysis represents an interesting extension of the analysis
in Aase and Gjesdal (2018), who treat the one period model. In that model
the intensity is graphically represented as a decreasing, convex function of
the decision variable κ (but there is no time development).
7
Here and in most of the numerical computations we do not go beyond two rounds in
the iterations indicated above.

27
 Fig. 4: The insider’s trading intensities as functions of κ.
In Figure 4 we present two graphs of β(t, κ) in the present continuous
model as a function of κ for given t, for two different points in time: t =
5 and t = 9, where the upper curve is for t = 9. These are also seen to be
decreasing. Unlike for the one-period model, these graphs are concave for
small values of κ and then becomes convex as κ increases. That these graphs
are decreasing in κ is in accordance with the results from the one period
model: The insider trades more softly the more the market maker perturbs
the price.

7.3 The variance function of the order flow y

We have already indicated a graph, Figure 1, illustrating the covariance func-
tion of the order flow. Here we study the time development of the variance
function V (t), as this quantity enters many of the key expressions in this
theory. The base case parameters are the same as before.

Fig. 5: The variance function V (t, κ) as a function of t.

The variance of the total order flow yt is seen to increase sharply in the

28
beginning, and then flattens out at around t = 3 when κ = 0.24. It is well
known from empirical studies that the volatility of the price increases as more
relevant information enters the market, since this causes trade to increase.
A reasonable model of insider trading should reflect this, and for our model
this is illustrated in Figure 5.
However, note that these variances decrease with κ for any given point t
in time. This is illustrated in Figure 6 for t = 1, 5, 9 as κ run from 0 to 0.5.
The more the market maker trades fees, the less the insider trades and the
lower the variance of the order flow yt .

Fig. 6: The variance function V (t, κ) as a function of κ.

The upper curve in Figure 6 corresponds to t = 9, then t = 5 and the
lowest curve is for t = 1.

7.4 The two parties’ net profits

Now we come to the important part, namely the net profit functions of the
two key participants as functions of time. These are illustrated in figures 7
and 8.
In Figure 7 we consider the dynamics of the two net profits as functions of
of time, for a given value of the parameter κ. We notice that both profit func-
tions naturally start out low and then increase with time. When κ = 0.045
the insider’s profit function cross the market maker’s profit from below at
around t = 7, and then ends up as the highest of the two at the final time
T = 108 .

8
The profit function of the insider is computed at discrete times only, due to the large
number of computations required for a continuous plot.

29
 Fig. 7: The two net profits as functions of time. κ = 0.045.
In Figure 8 however, where κ = 0.07, the market maker’s profit is seen to
dominate from the start, and also ends up highest at the final time T = 10.
When the market maker further increases κ, her profit function will increase
for each value of t, while the profit of the insider will decrease for each t
compared to the levels in Figure 8, (but will still be an increasing function
of t for each given value of κ, as in the figures 7 and 8). As a consequence,
the market maker outperforms the perfectly well informed insider from about
κ = 0.06 onwards.

Fig. 8: The two net profits as functions of time. κ = 0.07.

7.5 When does the market maker’s profit dominate?

We now illustrate the regulator’s problem through some graphs of the market
observable quantity rv. In the first figure we plot rv(t, κ) as functions of κ
for some given values of time t.

30
 Fig. 9: rv(t, κ) as a function of κ when t = 0.01, 3.6 and 9.0.
The lowest curve in Figure 9 corresponds to t = 0.01, the next lowest to
t = 9.0 and the highest curve to t = 3.6. To interpret this figure, imagine
that the regulator uses the rule that rv should correspond to less than 15%
increase from the fiduciary, ideal situation. This means that the market
maker should not increase κ beyond 0.07, at least based on the the three
time points in this illustration.
Since this line of reasoning is valid only for some particular values of t,
in the next figure we study rv’s time development:

Fig. 10: rv(t, κ) as a function of t when κ = 0.07.

Figure 10 illustrates the time picture of rv(t, κ) for κ = 0.07, for t ∈ [0, T ]
(the curved graph). The horizontal line tangent to the curve rv(t, 0.07) is
at level 1.15, corresponding to 15% maximal deviation from fiduciary trade.
From the figure it follows that the regulator will keep the market open all
the time.

31
 Fig. 11: rv(t, κ) as a function of t when κ = 0.09.
Figure 11 illustrates the same time picture of rv(t, κ) as in the previous
figure, but now for κ = 0.09. With a 15%-regulation rule still in charge, the
market would then be shut down and the stock suspended at around t = 1.2,
but allowed to reopen at around t = 6.5, and then kept open for the rest of
the period. Accordingly, the market maker would be wise to consider a lower
value of κ to keep her reputation as a decent professional.
On the other hand, if the regulator used a 21%-rule, the market would
stay open all the time for κ = 0.09.
The manner in which we find the inverse map from rv to κ is seen to
proceed as follows: For a given value of rv ∗ , say rv ∗ = 1.15, we solve in t and
κ the following inequality: supκ (supt rv(t, κ)) ≤ d∗ . Then the ’optimal’ value
of κ, call it κ∗ , is the one which satisfies this inequality with equality sign:
supt rv(t, κ∗ ) = d∗ . In Table 2 we illustrate this connection for some values
of rv ∗ .

Cont. model: t = 9
rv ∗ 1.03 1.05 1.08 1.15 1.21
κ∗ .025 .035 .045 .070 .090
pM (9, κ∗ ) .037 .049 .060 .087 .100
pI (9, κ∗ ) .126 .118 .100 .055 .045
ι(9, κ∗ ) .88 .83 .64 .45 .30

Table 2: The connection between κ∗ and rv ∗ with associated net profits.

In Table 2 the insider has the highest net profits, denoted pI (t, κ), for
lower values of rv ∗ , and this profit is decreasing with k ∗ . For larger values
of rv ∗ , the market maker’s profit is the largest of the two. The last row in
Table 2 illustrates the informativeness ι(t, κ) in the market as a function of κ

32
when t = 9. It decreases as κ increases. Distorting prices is not informative
to the other market participants.
In the above illustrations in the last two figures (figures 10 and 11) it
is the market maker who has the highest profits of the two parties. This
highlights one of the the main ideas in this paper: In real life we know that
market makers actually do not set prices in an entirely fiduciary manner,
but rather determine prices in such a way that they make money. In the
introduction we explained why this behavior is possible and likely to take
place. This is in line with observed behavior in several financial markets, in
particular those of the over-the-counter type that we have in mind.
For a modest fee conditional on the order flow, the market maker is able to
obtain a profit of the order of the magnitude, or even better than, a perfectly
informed insider, showing, among other things, the advantage of observing
the order flow. This, we conjecture, may be one explanation why so much
money tends to end up in the financial sector of the economy.

8 Suggestions for further research

In the discrete time paper of Aase and Gjesdal (2017) a situation is analyzed
where the market maker has private information as well. This could also
be of interest to analyze in the present setting. Thee is supposed to be no
information flow between the market making department and the investment
department of large financial institutions. But these ’Chinese Walls’ - as
they are known as - may not be entirely ’watertight’. Compared to the
model where the market maker has no privileged information, several new
insights were obtained in the above mentioned paper. As expected the well
informed insider’s profit diminishes since his informational advantage has
shrunk. This is because the market maker acts non-strategically with respect
to his private information, and allows this information to be reflected into the
price. Interestingly the informed market maker may then increase his fees
without being detected by the regulator, which in turn increases his profits.
Finally, we could analyze the situation where the market maker’s infor-
mation is public, which can be used to determine the effects of information
asymmetry. One major difference, discovered in the above mentioned paper,
is that the insider then trades harder when information is public, increasing
his own profit and that of the market maker. As another consequence, the
market is more efficient. Price distortions are now smaller as a function of

33
the price perturbation parameter allowing the market maker to make higher
profits. Thus it pays for him to share his private information with the insider,
but the noise traders are then even worse off. This analysis also throws some
new light on one ’positive’ side of insider trading (aside from the obvious
negative aspects which we do not dwell on here); all information arriving to
the market generally has a positive effect on efficiency.
In particular, price volatility is shown to increase with informed mar-
ket maker in the one period model. This is an important aspect of the
effect of privileged information on security prices, which may explain the
price/dividend puzzle, a feature that should be extend to the multiperiod
model (the world is, after all, time continuous).

9 Conclusions
The dynamic auction model of Kyle (1985) is studied, allowing market mak-
ers to maximize profit within regulatory limits by charging time varying,
stochastic fees. This has several implications for the equilibrium, the most
important being that by perturbing the price by a relatively modest amount,
the market maker is able to obtain a profit of the order of magnitude, and
even higher than, a perfectly informed insider.
The dynamic aspects of this are analyzed in the paper, and illustrated
numerically by examples. The analytical challenge turned out to be the
determination of the optimal trading intensity of the insider, when the market
maker perturbs the price. The solution was presented in Theorem 1 in Section
4. Based on this result, we were able to discuss a wide variety of problems,
like finding the profits of the two parties, the insider and the market maker,
the stochastic properties of the order flow, and the informativeness in the
market, all quantities as a functions of time and price perturbations. We also
indicated how a regulator can monitor the market by observing a measure
of relative price volatility, which compares to the corresponding measure
with fiducial price setting. This gives a convenient connection between price
volatility and price perturbation.
Our analysis indicates why speculative prices are more volatile than pre-
dicted by fundamentals.

34
10 Appendix 1.
Proof of Theorem 1 by Variational Calcu-
lus.
We now want to solve problem (4.2) by use of directional derivatives, or cal-
culus of variations. Towards this end, let A be the family of all continuously
differentiable functions β : [0, T ] → R such that
Z t 2
βs
2
ds < ∞ for all t < T.
0 σs

By this method we choose an arbitrary function ξt ∈ A, a sufficiently rich

set in this regard, and define the real function g by

g(x) = J I (k, β + xξ); x ∈ R.

Then, assuming β is optimal, g is maximal at x = 0 and hence the first order

condition of maximality at β is
d I
0 = g 0 (0) =J (k, β + xξ)|x=0 =
dx
 Z T Z t
(βs + xξs )2 −1

d 2 2
σ 1 + σṽ ds (βt + xξt )dt |x=0 +
dx ṽ 0 0 σs2
Z T  Z t  
d 2 −2 0t ks (βs +xξs )ds
R
2 2 0s kr (βr +xξr )dr
R
(βt + xξt )kt e σs e ds dt |x=0 =
dx 0 0

Z T Z T
βt
(10.1) γt (β) − 2 γs2 (β)βs ds 2 ξt dt+
0 t σt
Z T Rt
Z t Rs
kt2 e−2 0 ks βs ds σs2 e2 0 kr βr dr ds ξt dt+


0 0
Z T Rt
Z t Z t Rr
2 −2 0 kr βr dr
βt kt e (−2 ks ξs ds) σr2 e2 0 ku βu du dr)dt+
0 0 0
Z T Rt
Z t Rs
Z s
βt kt2 e−2 0 kr βr dr σs2 e2 0 kr βr dr (2 kr ξr dr)ds)dt = 0, ∀ξ ∈ A.
0 0 0

The second line on the left-hand side of (10.1) follows just as in Aase et. al
(2012c), which presents a simple proof of the case k = 0.

35
 Consider the last two lines, and start with the third integral in (10.1).
By changing the order of integration between s and t, we obtain
Z T Z t Z t
2 −2 0t kr βr dr
R Rr
βt kt e (−2 ks ξs ds) σr2 e2 0 ku βu du dr)dt =
0 0 0
Z T Z T Rt
Z t Rr
βt kt2 e−2 kr βr dr
( σr2 e2 0 ku βu du dr)dt ks ξs ds =


−2 0

0 s 0
Z T Z T Rs
Z s Rr
βs ks2 e−2 kr βr dr
σr2 e2 ku βu du


−2 0 ( 0 dr)ds kt ξt dt.
0 t 0

Next consider the fourth and last integral in (10.1). The inner integral given
by Z Z t Rs s
σs2 e2 0 kr βr dr
(2 kr ξr dr)ds)dt,
0 0
can be rewritten as
Z t Z t Rr
2 ( σr2 e2 0 ku βu du
dr)ks ξs ds.
0 s

This we now insert in the fourth term in (10.1), which gives

Z T Z T Rt
Z s Rr
2 βs ks2 e− 0 ku βu du
( σr2 e2 0 ku βu du dr)ds)kt ξt dt.
0 t t

Putting all this together, the first order condition now takes the form

d I
(10.2) 0 = J (k, β + xξ)|x=0 =
dx
Z T Z T Z t
βt 2 −2 0t ks βs ds
R Rt
γt (β) − 2( 2
γs (β) βs ds) + kt e σs2 e−2 0 kr βr dr ds
0 t σt 0
Z T Rs
Z s Rr
− 2kt βs ks2 e−2 0 kr βr dr ( σr2 e2 0 ku βu du dr)ds
t 0
Z T Rt
Z s 
2 2 0r ku βu du
R
2 −2 0 ku βu du
+ 2kt βs ks e ( σr e dr)ds ξt dt, ∀ξ ∈ A.
t t

36
Thus we conclude that
βt T
Z Z t
2 −2 0t ks βs ds
R Rt
(10.3) γt (β) = 2 2
γs (β) βs ds + kt e σs2 e−2 0 kr βr dr ds
σt t 0
Z T Rs
Z s Rr
− 2kt βs ks2 e−2 0 kr βr dr ( σr2 e2 0 ku βu du dr)ds
t 0
Z T Z s
2 −2 0t ku βu du
R Rr
+ 2kt βs ks e ( σr2 e2 0 ku βu du dr)ds.
t t

Accordingly
t
σt2  Rt
Z Rt
(10.4) βt = RT γt (β) − kt2 e−2 0 ks βs ds
σs2 e−2 0 kr βr dr
ds
2 t
γs (β)2 β s ds 0
Z T Rs
Z t 
2 2 0r ku βu du
R
− 2kt βs ks2 e−2 0 kr βr dr
( σr e dr)ds .
t 0

Hence, the optimal trading intensity of the insider, βt , t ∈ [0, T ], is given by

the integral equation (10.4).
This may be simplified by using the expression for the variance process
V (t) in (3.13). The result is

T
σt2  Z Rs

(10.5) βt = RT γt (β) − V (t) kt2 + 2kt βs ks2 e−2 t kr βr dr
ds .
2 t
γs (β)2 βs ds t

which is equation(4.3) in Theorem 1. 

11 Appendix 2.
Optimization via Pontryagin and Nash.
We start from the system expressed in terms of (yt , mt , γt ) in (3.8)-(3.10),
which are
(11.1) dyt = (ṽ − mt − kt yt )βt dt + σt dBt ; y0 = 0
γt βt
(11.2) dmt = 2 [(ṽ − mt )βt dt + σt dBt ]; m0 = E[v]
σt
β 2γ 2
(11.3) dγt = − t 2 t ; γ0 = E[(ṽ − E[ṽ])2 ].
σt

37
 The performance functionals are
Z T Z T
M M
(11.4) J (k, β) := w0 + E( kt yt (kt yt + mt − ṽ)βt dt − yt2 dkt )
0 0
Z T
(11.5) J I (k, β) := w0I + E[(ṽ − ms − ks ys )2 ]βs ds.
0

Problem 11.1. We want to find a Nash equilibrium (kt∗ , βt∗ ) for the the two
performance functionals J M , J I . In other words, we want to find (determin-
istic) control processes kt∗ , βt∗ such that
(11.6) sup J M (kt , βt∗ ) = J M (kt∗ , βt∗ )
kt

and
(11.7) sup J I (kt∗ , βt ) = J I (kt∗ , βt∗ ).
βt

This is a stochastic differential game.

Recall our assumption
(11.8) kt = κ(T − t); 0 ≤ t ≤ T,
for some constant κ ∈ [0, K], where K is a fixed constant (in principal set by
the regulator).
Then the performance functionals take the following forms
(11.9)
Z T
M
J (κ, β) := w0M + E[ {κ(T − t)yt (κ(T − t)yt + mt − ṽ)βt + κyt2 }dt]
0
(11.10)
Z T
I
J (κ, β) := w0 + E[(ṽ − mt − κ(T − t)yt )2 ]βt dt.
0

To study Problem 3.1 we use the stochastic maximum principle. Thus we

define two Hamiltonians H M and H I by
H M (t, y, m, γ, κ, β, p, q) = −κ(T − t)y(v − m − κ(T − t)y)β + κy 2
γβ
+ (v − m − κ(T − t)y)βp1 + σt q1 + 2 (v − m)βp2
σt
2 2
γβ β γ
(11.11) + q 2 − 2 p3
σt σt

38
and

H I (t, y, m, γ, κ, β, p, q) = (v − m − κ(T − t)y)2 β + (v − m − κ(T − t)y)βp1

γβ γβ β 2γ 2
(11.12) + σt q1 + 2 (v − m)βp2 + q2 − 2 p3 .
σt σt σt

The BSDE’s for the adjoint processes (pM M

i , qi ); i = 1, 2, 3, associated to H
M

are

M 2 2
dp1 (t) = [κ(T − t)βt (v − mt − 2κ(T − t)yt ) − κ (T − t) yt βt +

2κyt + κ(T − t)βt pM M
1 (t)]dt + q1 (t)dBt ; 0≤t≤T

 M
p1 (T ) = 0

γt βt2 M
(
dpM M
2 (t) = −[−κ(T − t)yt βt − βt p1 (t) − σt2 2
p (t)]dt + q2M (t)dBt ; 0 ≤ t ≤ T
pM
2 (T ) =0

βt2 (v−mt ) M 2βt2 γt M

(
βt M
dpM
3 (t) = −[ σ2
p2 (t) + q (t)
σt 2
− σt2
p3 (t)]dt + q3M (t)dBt ; 0 ≤ t ≤ T
t

pM
3 (T ) =0

The BSDE’s for the adjoint processes (pIi , qiI ); i = 1, 2, 3, associated to H I

are

I I
dp1 (t) = −[2(v − mt − κ(T − t)yt )(−κ(T − t))βt − κ(T − t)βt p1 (t)]dt

+q1I (t)dBt ; 0 ≤ t ≤ T

 I
p1 (T ) = 0

γt βt2 I

I I
dp2 (t) = −[2(v − mt − κ(T − t)yt )(−βt ) − βt p1 (t) −
 σt2 2
p (t)]dt
+q2I (t)dBt ; 0 ≤ t ≤ T

 I
p2 (T ) = 0

2βt2 γt I
(
βt I
dpI3 (t) = −[ v−m
σ2
t 2 I
βt p2 (t) + q (t)
σt 2
− σt2
p3 (t)]dt + q3I (t)dBt ; 0 ≤ t ≤ T
t

pI3 (T ) =0

39
 According to the maximum principle for stochastic differential games (see
e.g. [20] , Theorem 2.1 and Theorem 2.3) the problem of finding a Nash equi-
librium for the two performances J M (κ, β), J I (κ, β) can (under some condi-
tions) be reduced to finding a Nash equilibrium for the two Hamiltonians
H M , H I . Thus we proceed to maximize H M (κ, β) with respect to κ for each
given β, and then to maximize H I (κ, β) with respecty to β for each κ:
For each t the map

κ 7→ H M (t, yt , mt , γt , κ, pM (t), q M (t))

is convex and therefore it achieves its maximum κ = κ̂ at the boundary, i.e.

when
κ̂ = 0 or κ̂ = K.
Here K is the maximum allowed by the regulator.
The map
β 7→ H I (t, yt , mt , γt , κ, β, pI (t), q I (t))
has a critical point when

(11.13)
γt
 
(v − mt − κ(T − t)yt )2 + (v − mt − κ(T − t)yt )pI1 (t) + q (t)
σt 2
σt2
β = β̂(t) = −
2[γt (v − mt )pI2 (t) − γt2 p3 (t)]

We conclude the following:

Theorem 11.2. Suppose (κ∗ , β ∗ ) is a Nash equilibrium for Problem 3.1.

Then
κ∗ = 0 or κ∗ = K,
and the optimal β ∗ is given in feedback form by

(11.14)
γt∗ I ∗
∗
(v − m∗t − κ∗ (T − t)yt∗ )2 + (v − m∗t − κ∗ (T − t)yt∗ )(pI1 )∗ (t) + (q ) (t)
σt 2
β (t) = γ2 γt∗
2[ σt2 (pI3 )∗ (t) − σt2
(v − m∗t )(pI2 )∗ (t)]
t

where yt∗ , m∗t , γt∗ , (pI1 )∗ (t), (pI2 )∗ (t), (pI3 )∗ (t), (q2I )∗ (t) are the system values cor-
responding to the controls κ∗ , β ∗ .

40
12 Appendix 3.
The Bellman approach.
It may be instructive to see what the dynamic programming approach gives
in the present situation. In particular, this may throw some light on the
interpretations of the adjoint variables in Theorem 3.2. In doing so, we take
into account our previous remarks made just prior to equation (2.4) in Section
2.2, which tells us to focus on the insider’s profit only, since the market maker
does not act strategically, he only trades ’fees’.
Let us for short use the notation xt = (yt , mt , γt ) for the system. The
performance functional is given in (3.5), and the maximal profit of the insider
is
hZ T 
2
I I

(12.1) J (x) = w0 + supβ E ṽ − ms − ks ys βs ds .
0

With J I (x, t) equal to the optimal wealth remaining at time t in state x, the
Bellman equation can be written
n o
(12.2) supβ (v − mt − kt yt )2 βt + Lβ (J I (x, t)) = 0,

where

(12.3)
β I ∂J I (x, t) ∂J I (x, t) γt βt ∂J I (x, t)
L (J (x, t)) = +(v−mt −kt yt )βt + 2 (v−mt )βt
∂t ∂y σt ∂m
2 2 I 2 I 2 I 2 I
β γ ∂J (x, t) ∂ J (x, t) ∂ J (x, t) ∂ J (x, t)
− t 2t + σt 2
+ σt 2
+ 2σt .
σt ∂γ ∂y ∂m ∂y∂m

Let us first address the maximization problem in (12.2). The first order
condition in βt can be written

∂J I (x, t) γt ∂J I (x, t)
(v − mt − kt yt )2 + (v − mt − kt yt ) + 2βt 2 (v − mt )
∂y σt ∂m

γt2 ∂J I (x, t)
−2βt = 0.
σt2 ∂γ

41
This gives the optimal βt∗ in terms of the function J I (x, t) (i.e., its partial
derivatives) as follows
I
(v − mt − κ(T − t)yt )2 + (v − mt − κ(T − t)yt ) ∂J ∂y(x,t)
(12.4) βt∗ = γ2 I I
.
2[ σt2 ∂J ∂γ(x,t) − γt
σt2
(v − mt ) ∂J∂m
(x,t)
]
t

It remains to determine the function J I (x, t).

Comparing this expression for the optimal trading intensity of the insider
with the corresponding expression in (3.14) derived using the stochastic max-
imum principle for stochastic differentiable games, we notice that the adjoint
variables pIi (t), i = 1, 2, 3 and q2I (t) can be expressed as follows
∂J I (x, t) ∂J I (x, t)
(pI1 )∗ (t) = , (pI2 )∗ (t) = ,
∂y ∂m
I ∗ ∂J I (x, t)
(p3 ) (t) = , (q2I )∗ (t) = 0.
∂γ
These relationships give us more insight into how to interpret these adjoint
variables, namely for the pIi (t)’s as marginal profits with respect to the state
variables of the problem at any time t ∈ (0, T ). It also says that the adjoint
variable (pI2 )∗ (t) has no diffusion term, so this is a finite variation process.
Notice that it is not normally the case that the adjoint variables in the
stochastic maximum principle can be interpreted as ’shadow prices’ as we
can here. When the state variables have different volatilities, or driven by
different Brownian motions, this will in general no longer be true (see e.g.,
Yong and Zhou (1999)).
RT
The insider’s indirect utility function J I (x, t) = t βs (γs (β)+ks2 Vs )ds, can
be written as a function of the state x variable and time t as indicated by this
notation, and from a conjectured functional form we may attempt to solve
the Bellman equation, and proceed to solution for the trading intensity β.
The result of this can in its turn be used to address the problem of Appendix
2. However, here we choose to stop and leave this for future research.

13 Appendix 4.
A connection to filtering theory.
The results of Section 3.1 can alternatively be derived using filter theory
as follows: We first consider the process y for k = 0. Then dyt = (ṽ −

42
E(ṽ|Fty )βt + σt dBt . From filtering theory (see Allinger and Mitter (1981))
we then know that y generates the same filtration as ŷ, i.e., Ftŷ = Fty , and
that ỹ defined by dỹt := σ1t dyt := dbt is a Brownian motion with respect to
the information filtration Fty . 9
Employing this result to our situation when k 6= 0, , we obtain that
1
{dyt + kt βt yt dt} := dbt
σt
for an Fty -Brownian motion bt . We may express the total order process y as
follows
dyt = −kt βt yt dt + σt dbt .
We now employ standard results for Gaussian processes to find µt := E(yt )
and V (t) := E(yt2 ) for all t ∈ [0, T ]. Using Karatzas and Schreeve (1985), we
have that µ(t) = E(yt ) = 0 for all t provided y0 = 0, and the following first
order non-homogeneous ordinary linear differential equation for the variance
V (t) = E(yt2 ),
dV (t)
= −2kt βt V (t) + σt2 , V (0) = 0
dt
which has the solution
Z t
−2 0t ks βs ds
R Rs
2
(13.1) V (t) = E(yt ) = e σs2 e2 0 kr βr dr ds.
0

This is (3.13).

Acknowledgments

References
[1] Aase, K. K. and F. Gjesdal (2017). ”Insider trading with non-fiduciary
market makers.”. (Last version 2018) Discussion Paper FOR 8, Norwe-
gian School of Economics.

[2] Aase, K. K. and B. Øksendal (2018). ”Strategic Insider Trading in Con-

tinuous Time: A New Approach.” Discussion Paper FOR, Norwegian
School of Economics.
9
The result by Allinger and Mitter proved a long-standing conjecture by Kailath.

43
 [3] Aase, K., Bjuland, T. and Øksendal, B. (2012a). ”Partially informed
liquidity traders.” Mathematics and Financial Economics 6, 93-104.

[4] Aase, K., Bjuland, T. and Øksendal, B. (2012b). ”Strategic Insider

Trading Equilibrium: A Filter Theory Approach.” Afrika Matematika
23(2),145-162.

[5] Aase, K., Bjuland, T. and Øksendal, B. (2011). ”An anticipative linear
filtering equation.” Systems & Control Letters 1-4.

[6] Admati, A. R. and Pfleiderer, P. (1988). ”A Theory of Intraday Patterns:

Volume and Price Variability.” The Review of Financial Studies 1, 1, 3–
40.

[7] Allinger, D. F. and Mitter, S. K. (1981). ”New Results on the Innova-

tions Problem for Non-Linear Filtering”. Stochastics 4, 339–348.

[8] Back, K. (1992). ”Insider Trading in Continuous Time”. The Review of

Financial Studies Vol. 5, No 3, 387-409.

[9] Biagini, F., and Øksendal, B. (2005). ”A general stochastic calculus

approach to insider trading”. Appl. Math. Optim. 52, 167–181.

[10] Davis, M. H. A. (1977). Linear Estimation and Stochastic Control.

Chapman and Hall.

[11] Davis, M. H. A. (1984). Lectures on Stochastic Control and Nonlinear

Filtering.. Tata Institute of Fundamental Research, Bombay.

[12] Eide, I. B. (2007). ”An equilibrium model for gradually revealed asym-
metric information”. Preprint, University of Oslo 6/2007.

[13] Glosten, L. R., and Milgrom, P. R. (1985). ”Bid, Ask and Transaction
Prices in a Specialist Market with Hetrogeneously Informed Traders”.
Journal of Financial Economics, 14, 71–100.

[14] Grossman, S. J., and Stiglitz, J. E. (1980). ”On the Impossibility of

Informationally Efficient Markets”. American Economic Review, 70 ,
393–408.

44
[15] Holden, C. W. and Subrahmanyam, A. (1992). ”Long-Lived Private In-
formation and Imperfect Competition”. The Journal of Finance, XLVII,
1 247–270.

[16] Kallianpur, G. (1980). Stochastic Filtering Theory. Springer.

[17] Kalman, R. E. (1960). ”A new approach to linear filtering and prediction

problems”. J. Basic Engineering D 82, 35-45.

[18] Karatzas I., and S. Schreve (1988). Brownian motion and stochastic
calculus. New York: Springer-Verlag.

[19] Kyle, A. S. (1985). ”Continuous Auctions and Insider Trading”. Econo-

metrica Vol.53, No. 6, 1315–1336.

[20] Liptser, R. S. and Shiryaev, A. N.: Statistics of Random Processes II.

Springer 1978.

[21] Øksendal, B. (2013). ”Stochastic Differential Equations.” 6th Edition.

Springer

[22] Russo, F., and Vallois, P. (1993). ”Forward, backward and symmetric
stochastic integration”. Probab. Theory Related Fields 97, 403–421.

[23] Russo, F. and Vallois, P. (1995). ”The generalized covariation process

and Itô formula”. Stoch. Process. Appl. 59, 81–104.

[24] Russo, F. and Vallois, P. (2000). ”Stochastic calculus with respect to

continuous finite quadratic variation processes”. Stoch. Stoch. Rep. 70,
1–40.

[25] Tucker, H. G. (1967). A graduate course in probability. Academic Press,

New York and London.

[26] Yong, J. and X. Y. Zhou (1999). ”Hamiltonian Systems and the HJB
Equations”. Springer-Verlag, New York, Inc.

45