Nonlinear Models in Option Pricing an Introduction

Matthias Ehrhardt
WeierstraInstitut fr Angewandte Analysis und Stochastik, Mohrenstr. 39, 10117 Berlin, Germany
Abstract. Nonlinear BlackScholes equations have been increasingly attracting interest over
the last two decades, since they provide more accurate values by taking into account more realistic
assumptions, such as transaction costs, risks from an unprotected portfolio, large investors prefer-
ences or illiquid markets, which may have an impact on the stock price, the volatility, the drift and
the option price itself.
This book consists of a collection of contributed chapters of wellknown outstanding scientists
working successfully in this challenging research area. It discusses concisely several models from
the most relevant class of nonlinear BlackScholes equations for European and American options
with a volatility depending on different factors, such as the stock price, the time, the option price
and its derivatives. We will present in this book both analytical techniques and numerical methods
to solve adequately the arising nonlinear equations.
The purpose of this book is to give an overview on the current state-of-the-art research on
nonlinear option pricing. The intended audience is on the one hand graduate and Ph.D. students
of (mathematical) nance and on the other hand lecturer of mathematical nance and and people
working in banks and stock markets that are interested in new tools for option pricing.
1 Introduction
Nonlinear models im mathematical nance are becoming more and more important since
they take into account effects like the presence of transaction costs, feedback and illiquid
market effects due to large traders choosing given stock-trading strategies, imperfect repli-
cation and investors preferences and risk from unprotected portfolios.
Due to transaction costs, illiquid markets, large investors or risks from an unprotected
portfolio the assumptions in the classical BlackScholes model become unrealistic and
the model results in strongly or fully nonlinear, possibly degenerate, parabolic diffusion
convection equations, where the stock price, volatility, trend and option price may depend
on the time, the stock price or the option price itself.
In this chapter we will be concerned with several models from the most relevant class
of nonlinear BlackScholes equations for European and American options with a volatility
depending on different factors, such as the stock price, the time, the option price and its
derivatives, where the nonlinearity results from the presence of transaction costs.
In the following sections we will give a short introduction to option pricing.
1
2 Matthias Ehrhardt
2 Financial Derivatives
The interest in pricing nancial derivatives among them in pricing options arises from
the fact that nancial derivatives, also called contingent claims, can be used to minimize
losses caused by price uctuations of the underlying assets. This process of protection is
called hedging. There is a variety of nancial products on the market, such as futures,
forwards, swaps and options. In this introductory chapter we will focus on European and
American Call and Put options.
Denition 2.1 A European Call option is a contract where at a prescribed time in the future,
known as the expiry or expiration date T (t = 0 means today), the holder of the option
may purchase a prescribed asset, known as the underlying asset or the underlying S(t), for
a prescribed amount, known as the exercise or strike price K. The opposite party, or the
writer, has the obligation to sell the asset if the holder chooses to buy it.
At the nal time T the holder of the European Call option will check the current price of the
underlying asset S := S(T). If the price of the asset is greater than the strike price, S K,
then the holder will exercise the Call and buy the stock for the strike price K. Afterwards,
the holder will immediately sell the asset for the price S and make a prot of V = S K.
In this case the cash ow, or the difference of the money received and spent, is positive and
the option is said to be in-the-money. If S = K, the cash ow resulting from an immediate
exercise of the option is zero and the option is said to be at-the-money. In case S K, the
cash ow is negative and the option is said to be out-of-the-money. In the last two cases the
holder will not exercise the Call option, since the asset S can be purchased on the market
for K or less than K, which makes the Call option worthless. Therefore, the value of the
European Call option at expiry, known as the pay-off function, is
V (S, T) = (S K)
+
,
with the notation f
+
= max(f, 0).
Denition 2.2 Reciprocally, a European Put option is the right to sell the underlying asset
S(t) at the expiry date T for the strike price K. The holder of the Put may exercise this
option, the writer has the obligation to buy it in case the holder chooses to sell it.
The Put is in-the-money if K S, at-the-money if K = S and out-of-the-money if K S.
The pay-off function for a European Put option is therefore
V (S, T) = (K S)
+
.
The pay-off functions for the European Call and Put option are plotted in Fig, 1 from the
perspective of the holder. This perspective is called the long position. The perspective of
the writer, or the short position, is reversed and can be seen when the pay-off functions in
Fig. 1 are multiplied by 1. That means that the writer of a European Call option is taking
the risk of a potentially unlimited loss and must carefully design a strategy to compensate
for this risk [27].
Nonlinear Models in Option Pricing an Introduction 3
V
S
European Call
0 K
V
S
European Put
0 K
K
Figure 1: Pay-off functions for European options with a strike price K.
While European options can only be exercised at the expiry date T, American options
can be exercised at any time until the expiration. Since an American option includes at least
the same rights as the corresponding European option, the value of an American option
V
am
can never be smaller than the value of a European option V
eur
, i.e.
V
am
V
eur
.
Whether the values are equal depends on the dividend yield q, which describes the percent-
age rate of the returns on the underlying asset. Assuming that the underlying stock S pays
no dividends, the values of a European and an American Call option are equal if all the
other parameters remain the same (for details see [13, 34]). In case of an American Put
option without dividend payments it can often be advantageous to exercise it before expiry,
so that the values of a European and an American Put can differ substantially.
In the presence of a continuous dividend payment the fair price V (S, 0) of both an
American Call and Put option is greater than the value of a European Call or Put, see Fig. 2.
K

S
V
(
S
,
0
)
European Call
American Call
PayOff V(S,T)
(a) American vs. European Call with dividends.
K
K
S


V
(
S
,
0
)
European Put
American Put
Payoff V(S,T)
(b) American vs. European Put option.
Figure 2: Schematical values of American vs. European options at t = 0.
4 Matthias Ehrhardt
Furthermore, it should be mentioned that the value of a Call option on an underlying
without a dividend payment is always greater than the value of a Call option on an under-
lying with a dividend payment for both European and American options. For European
and American Put options on an underlying without a dividend payment the value is less
than on an underlying with a dividend payment. The inuence of a dividend payment is
summarized in Fig. 3.
Options, whose pay-offs only depend on the nal value of the underlying asset, are
called vanilla options. Options, whose pay-offs depend on the path of the underlying asset,
are called exotic or path-dependent options. Examples are Asian, Barrier and lookback
options. In this chapter, we will be solely concerned with plain vanilla European and Amer-
ican options.
K
S
V
(
S
,
0
)


q=0
q=0.2
q=0.4
Payoff V(S,T)
(a) European Call option with dividend yields q.
K
K
S
V
(
S
,
0
)


q=0
q=0.2
q=0.4
Payoff V(S,T)
(b) European Put option with dividend yields q.
K
V
(
S
,
0
)

S
q=0
q=0.2
q=0.4
Payoff V(S,T)
(c) American Call option with dividend yields q.
K
K
S
V
(
S
,
0
)


q=0
q=0.2
q=0.4
Payoff V(S,T)
(d) American Put option with dividend yields q.
Figure 3: The inuence of a dividend yield.
Nonlinear Models in Option Pricing an Introduction 5
3 Linear BlackScholes Equations
Option pricing theory has made a great leap forward since the development of the Black
Scholes option pricing model by Fischer Black and Myron Scholes in [3] in 1973 and
previously by Robert Merton in [25]. The solution of the famous (linear) BlackScholes
equation [9]
0 = V
t
+
1
2

2
S
2
V
SS
+rSV
S
rV, (1)
where S := S(t) > 0 and t (0, T), provides both an option pricing formula for a
European option and a hedging portfolio that replicates the contingent claim assuming that
[27]:
The price of the asset price or underlying asset S follows a Geometric Brownian
motion, meaning that if W := W(t) is a standard Brownian motion (see Appendix
A.6), then S satises the following stochastic differential equation (SDE):
dS = Sdt +SdW.
The trend or drift (measures the average rate of growth of the asset price), the
volatility (measures the standard deviation of the returns) and the riskless interest
rate r are constant for 0 t T and no dividends are paid in that time period.
The market is frictionless, thus there are no transaction costs (fees or taxes), the
interest rates for borrowing and lending money are equal, all parties have immediate
access to any information, and all securities and credits are available at any time and
any size. That is, all variables are perfectly divisible and may take any real number.
Moreover, individual trading will not inuence the price.
There are no arbitrage opportunities, meaning that there are no opportunities of in-
stantly making a risk-free prot ("There is no such thing as free lunch").
Under these assumptions the market is complete, which means that any derivative and any
asset can be replicated or hedged with a portfolio of other assets in the market (see [31]).
Then, it is well-known that the linear BlackScholes equation (1) can be transformed into
the heat equation and analytically solved to price the option [33]. The derivation of the so-
lution can be found in [27], the formulae for the European Call and Put options are attached
in Appendix B.
For American options, in general, analytic valuation formulae are not available, except
for a few special types, which we are not going to address in this chapter. Those types are
Calls on an asset that pays discrete dividends and perpetual Calls and Puts meaning Calls
and Puts with an innite time to expiry [23].
6 Matthias Ehrhardt
4 Nonlinear BlackScholes Equations
It is quite easy to imagine that the restrictive assumptions mentioned in the previous Sec-
tion 3 are never fullled in reality. Due to transaction costs (cf. [2, 5, 24]), large investor
preferences (cf. [11, 12, 26]) and incomplete markets (cf. [30]) these assumptions are likely
to become unrealistic and the classical model results in strongly or fully nonlinear, possibly
degenerate, parabolic convectiondiffusion equations, where both the volatility and the
drift can depend on the time t, the stock price S or the derivatives of the option price V
itself.
In this chapter we will focus on several transaction cost models from the most rele-
vant class of nonlinear BlackScholes equations for European and American options with
a constant drift and a nonconstant modied volatility function

2
:=
2
(t, S, V
S
, V
SS
).
Under these circumstances (1) becomes the following nonlinear BlackScholes equation,
which we will consider for European options:
0 = V
t
+
1
2

2
(t, S, V
S
, V
SS
)S
2
V
SS
+rSV
S
rV, (2)
where dS = Sdt + SdW, S > 0 and t (0, T).
Studying the linear BlackScholes equation (1) for an American Call option would be
redundant, since the value of an American Call option equals the value of a European Call
option if no dividends are paid and the volatility is constant.
In order to make the model more realistic, we will consider a modication of the non-
linear BlackScholes equation (2) for American options, where S pays out a continuous
dividend qSdt in a time step dt:
0 = V
t
+
1
2

2
(t, S, V
S
, V
SS
)S
2
V
SS
+ (r q)SV
S
rV, (3)
where S follows the dynamics dS = ( q)Sdt + SdW, S > 0, t (0, T) and the
dividend yield q is constant.
In the mathematical sense the nonlinear BlackScholes equations (2) and (3) are called
convectiondiffusion equations. The second-order term
1
2

2
(t, S, V
S
, V
SS
)S
2
V
SS
is re-
sponsible for the diffusion, the rst-order term rSV
S
or (r q)SV
S
is called the convection
term and rV can be interpreted as the reaction term (see [27, 32]).
In the nancial sense, the partial derivatives indicate the sensitivity of the option price
V to the corresponding parameter and are called Greeks. The option delta is denoted by
= V
S
, the option gamma by = V
SS
and the option theta by = V
t
. For a detailed
discussion of this issue we refer to [19].
5 Terminal and Boundary Conditions
In order to nd a unique solution for the equation (2) we need to complete the problem by
stating the terminal and boundary conditions for both the European Call and Put option.
Nonlinear Models in Option Pricing an Introduction 7
Since American options can be exercised at any time before expiry, we need to nd
the optimal time t of exercise, known as the optimal exercise time. At this time, which
mathematically is a stopping time (see Appendix A.5), the asset price reaches the optimal
exercise price or optimal exercise boundary S
f
(t). This leads to the formulation of the
problem for American options by dividing the domain [0, [[0, T] of (3) into two parts
along the curve S
f
(t) and analyzing each of them (see Fig. 4). Since S
f
(t) is not known in
advance but has to be determined in the process of the solution, the problem is called free
boundary value problem [34].
hold exercise
S
0
T
t
S
f
(0) S
f
(T)
S
f
(t)
(a) American Call.
hold exercise
S
0
T
t
S
f
(0) S
f
(T)
S
f
(t)
(b) American Put.
Figure 4: Exercising and holding regions for American options.
For different numerical approaches, the free boundary problem for American options
can be reformulated into a linear complementary problem (LCP), a variational inequality
and a minimization problem [13]. The most simple treatment is the formulation as a free
boundary problem [8, 14].
Even though we will focus on Call options in this chapter, we state the conditions for
Put options for the sake of completeness.
5.1 European Call Option
The value V (S, t) of the European Call option is the solution to (2) on 0 S < ,
0 t T with the following terminal and boundary conditions:
V (S, T) = (S K)
+
for 0 S <
V (0, t) = 0 for 0 t T (4)
V (S, t) S Ke
r(Tt)
as S .
5.2 European Put Option
Reciprocally, the value V (S, t) of the European Put option is the solution to (2) on 0
S < , 0 t T with the payoff function for the Put as the terminal condition and the
8 Matthias Ehrhardt
boundary conditions:
V (S, T) = (K S)
+
for 0 S <
V (0, t) = Ke
r(Tt)
for 0 t T (5)
V (S, t) 0 as S .
5.3 American Call Option
For the American Call option the spatial domain is divided into two regions by the free
boundary S
f
(t), the stopping region S
f
(t) < S < , 0 t T, where the option is
exercised or dead with V (S, t) = S K and the continuation region 0 S S
f
(t),
0 t T, where the option is held or stays alive and (3) is valid under the following
terminal and boundary conditions (see Fig. 4(a)):
V (S, T) = (S K)
+
for 0 S S
f
(T)
V (0, t) = 0 for 0 t T
V (S
f
(t), t) = S
f
(t) K for 0 t T (6)
V
S
(S
f
(t), t) = 1 for 0 t T
S
f
(T) = max(K, rK/q).
For the sake of simplicity we will assume r > q in this chapter, and therefore we have
S
f
(T) = rK/q for the American Call.
The structure of the value of an American Call can be seen Fig. 5(a), where we notice
that the free boundary S
f
(t) determines the position of the exercise. The exercising and
holding regions are illustrated in Fig. 4(a).
5.4 American Put Option
The American Put option is exercised in the stopping region 0 S < S
f
(t), 0 t T
where it has the value V (S, t) = K S (see Fig. 4(b)). In the continuation region S
f
(t)
S < , 0 t T the Put option stays alive and (3) is valid under the following terminal
and boundary conditions:
V (S, T) = (K S)
+
for S
f
(T) S <
lim
S
V (S, t) = 0 for 0 t T
V (S
f
(t), t) = K S
f
(t) for 0 t T (7)
V
S
(S
f
(t), t) = 1 for 0 t T
S
f
(T) = min(K, rK/q).
Since we assumed that r > q, we have S
f
(T) = K for the American Put. In Fig. 5(b) one
can see how the free boundary S
f
(t) determines the structure of an American Put.
Nonlinear Models in Option Pricing an Introduction 9
V (S, t)
K
S
0
T
t
S
f
(0)
S
f
(T)
S
f
(t)
(a) American Call.
V (S, t)
K
K
S
0
T
t
S
f
(0)
S
f
(T)
S
f
(t)
(b) American Put.
Figure 5: Schematical values V (S, t) of American options.
6 Volatility Models
The essential parameter of the standard BlackScholes model, that is not directly observable
and is assumed to be constant, is the volatility . There have been many approaches to
improve the model by treating the volatility in different ways and using a modied volatility
function () to model the effects of transaction costs, illiquid markets and large traders,
which is the reason for the nonlinearity of (2) and (3). We will rst give a brief overview of
several volatility models and then focus on the volatility models of transaction costs.
The constant volatility in the standard BlackScholes model can be replaced by
the estimated volatility from the former values of the underlying. This volatility is
known as the historical volatility [13].
If the price of the option and the other parameters are known, which is e.g. the case
for the European Call and Put options (see Appendix B), then the implied volatility
can be calculated from those BlackScholes formulae. The implied volatility is the
value , for which (24) or (25) is true compared to the real market data. It can be
calculated implicitly via the difference between the observed option price V (fromthe
market data) and the BlackScholes formulae (24) or (25), where all the parameters
except for the implied volatility are taken from the market data (the stock price
S, the time t, the expiration date T, the strike price K, the interest rate r the dividend
rate q).
Considering options with different strike prices K but otherwise identical parameters,
we see that the implicit volatility changes depending on the strike price. If the implicit
volatility for a certain strike price K is less than the implicit volatility for both the
strike price greater and less than K, this effect is called volatility smile [22].
10 Matthias Ehrhardt
Replacing the constant volatility with the observed implicit volatilities at each stock
price and time leads to the term of the local volatility := (S, t). Dupire [7]
examined the dependencies and expressed the local volatility as a function of implicit
volatilities.
Hull and White [18] and Heston [15] developed a model, in which the volatility fol-
lows the dynamics of a stochastic process. This is known as the stochastic volatility.
The assumption, that each security is available at any time and any size, or that in-
dividual trading will not inuence the price, is not always true. Therefore, illiquid
markets and large trader effects have been modeled by several authors. In [11] Frey
and Stremme and later Frey and Patie [12] considered these effects on the price and
come up with the result
=

1 (S)SV
SS
, (8)
where the historical volatility, constant, (S) strictly convex function, (S) 1.
The function (S) depends on the pay-off function of the nancial derivative. For the
European Call option, Frey and Patie show that (S) is a smooth, slightly increasing
function for S K. Bordag and Chmakova [4] assumed that (S) is constant
and solve the problem (2) with the modied volatility (8) explicitly using Lie-group
theory (see also [6]).
As the main scope of this general overview chapter, we draw our attention in the sequel to
a more detailed description of several transaction cost models.
6.1 Transaction Costs
The BlackScholes model requires a continuous portfolio adjustment in order to hedge the
position without any risk. In the presence of transaction costs it is likely that this adjustment
easily becomes expensive, since an innite number of transactions is needed [23]. Thus, the
hedger needs to nd the balance between the transaction costs that are required to rebalance
the portfolio and the implied costs of hedging errors. As a result to this "imperfect" hedging,
the option might be over- or underpriced up to the extent where the riskless prot obtained
by the arbitrageur is offset by the transaction costs, so that there is no single equilibrium
price but a range of feasible prices.
It has been shown that in a market with transaction costs there is no replicating portfolio
for the European Call option and the portfolio is required to dominate rather than replicate
the value of the option (see [2]). Soner, Shreve and Cvitani c [29] proved that the minimal
hedging portfolio that dominates a European Call is the trivial one (hence holding one share
of the stock that the Call is written on), so that efforts have been made to nd an alternate
relaxation of the hedging conditions to better replicate the pay-offs of derivative securities.
Nonlinear Models in Option Pricing an Introduction 11
6.2 The model of Leland
Lelands idea [24] of relaxing the hedging conditions is to trade at discrete times, which
promises to reduce the expenses of the portfolio adjustment. He assumes that the transac-
tion cost ||S/2, where denotes the round trip transaction cost per unit dollar of the
transaction and the number of assets bought ( > 0) or sold ( < 0) at price S, is
proportional to the monetary value of the assets bought or sold. Now consider a replicating
portfolio with units of the underlying and the bond B (a certicate of debt issued by
a government or a corporation guaranteeing payment B plus interest by a specied future
date):
= S +B.
After a small change in time of the size t the change in the portfolio becomes
= S +rBt

2
||S, (9)
where S is the change in price S, so that the rst term represents the change in value,
the second term represents the bond growth in t time and represents the change in the
number of assets, so that the last term becomes the transaction cost due to portfolio change.
We apply Its lemma (see 23 in Appendix A.7) to the value of the option V := V (S, t)
and get
V = V
S
S + (V
t
+

2
2
S
2
V
SS
)t. (10)
Assuming that the option V is replicated by the portfolio , their values have to match at
all times and there can be no risk-free prot. With this no-arbitrage argument we get
= V.
Matching the terms in (9) and (10) we get = V
S
and
rBt

2
||S = (V
t
+

2
2
S
2
V
SS
)t. (11)
Leland shows that

2
||S =

2
2
Le S
2
|V
SS
|t, (12)
where Le denotes the Leland number, which is given by
Le =
_
2

t
_
, (13)
with t being the transaction frequency (interval between successive revisions of the port-
folio) and the round trip transaction cost per unit dollar of the transaction. Plugging (12)
and B = S = V SV
S
into the equation (11) becomes
rV rSV
S


2
2
Le S
2
|V
SS
| = V
t
+

2
2
S
2
V
SS
. (14)
12 Matthias Ehrhardt
Therefore, Leland deduces that the option price is the solution of the nonlinear Black
Scholes equation
0 = V
t
+
1
2

2
S
2
V
SS
+rSV
S
rV,
with the modied volatility

2
=
2
_
1 + Le sign(V
SS
)
_
, (15)
where represents the historical volatility and Le the Leland number. It follows from the
denition of the Leland number (13) that the more frequent the rebalancing (t smaller),
the higher the transaction cost and the greater the value of V .
It is known that V
SS
> 0 for European Puts and Calls in the absence of transaction costs.
Assuming the same behavior in the presence of transaction costs, equation (2) becomes
linear with an adjusted constant volatility
2
=
2
(1 + Le) >
2
.
Lelands model has played a signicant role in nancial mathematics, even though it has
been partly criticized by e.g. Kabanov and Safarian in [21], who prove that Lelands result
has a hedging error. The restriction of his model is the convexity of the resulting option
price V (hence V
SS
> 0) and the possibility to only consider one option in the portfolio.
Hoggard, Whalley and Wilmott studied equation (2) with the modied volatility (15) for
several underlyings in [17]. An extension to this approach to general pay-offs is obtained
by Avellaneda and Pars [1].
6.3 Barles and Soner
In [2] Barles and Soner derived a more complicated model by following the above utility
function approach of Hodges and Neuberger [16]. Consider the process of bonds owned
X(s) and the process of shares owned Y (s). Let the trading strategy
_
L(s), M(s)
_
be
a pair of nondecreasing processes with L(t) = M(t) = 0, which are interpreted as the
cumulative transfers, measured in shares of stock. L(s) is measured in shares from bond to
stock and M(s) is measured in shares fromstock to bond. Let (0, 1) be the proportional
transaction cost. The processes X(s) and Y (s) start with the initial values x and y, s
[t, T] and evolve according to
X(s) = x
_
s
t
S()(1 +) dL() +
_
s
t
S()(1 ) dM() (16)
and
Y (s) = y +L(s) M(s). (17)
The rst integral in (16) represents buying shares of stock at a price increased by the pro-
portional transaction cost, the second integral represents selling stock at a reduced price
of the transaction cost. In (17) we add the amount of the stocks bought and subtract the
amount for the stocks sold to the initial amount of stocks owned.
Nonlinear Models in Option Pricing an Introduction 13
According to the utility maximization approach of Hodges and Neuberger [16], the
price of a European Call option can be obtained as the difference between the maximum
utility of the terminal wealth when there is no option liability and when there is such a
liability. Following this approach, Barles and Soner considered two optimization problems.
Let the exponential utility function be
U() = 1 e

, R,
where > 0 is the risk aversion factor. The rst value function is the expected utility from
the nal wealth without any option liabilities taken over the transfer processes
V
1
(x, y, S(t), t) := sup
L(),M()
E[U
_
X(T) +Y (T)S(T)
_
],
the second one is the expected utility from the nal wealth assuming that we have sold N
European Call options taken over the transfer processes
V
2
(x, y, S(t), t) := sup
L(),M()
E[U
_
X(T) +Y (T)S(T) N(S(T) K)
+
_
].
Hodges and Neuberger postulate that the price of each option is equal to the maximal solu-
tion of the algebraic equation
V
2
(x +N, y, S(t), t) = sup
L(),M()
E[U
_
X(T) +N +Y (T)S(T)
N(S(T) K)
+
_
]
= sup
L(),M()
E[U
_
X(T) +Y (T)S(T)
_
]
= V
1
(x, y, S(t), t),
which means that the option price equals the increment of the initial capital at time t that
is needed to cope with the option liabilities arising at T. By a linearity argument selling N
options with risk aversion factor of yields the same price as selling one option with risk
aversion factor N. This leads to performing an asymptotic analysis as N . Hence,
we consider
U() = 1 e
N
and
=
1
N
.
Then, we have
U

() = 1 e

, R.
Our optimization problems become
V
1
(x, y, S(t), t) = 1 inf
L(),M()
E[e

(X(T)+Y (T)S(T))
]
14 Matthias Ehrhardt
and
V
2
(x, y, S(t), t) = 1 inf
L(),M()
E[e

(X(T)+Y (T)S(T)(S(T)K)
+
)
].
For analysis simplication Barles and Soner dene z
1,2
: R (0, ) (0, T) R by
V
1
(x, y, S(t), t) = 1 e

_
x+yS(t)z
1
(y,S(t),t)
_
and
V
2
(x, y, S(t), t) = 1 e

_
x+yS(t)z
2
(y,S(t),t)
_
.
Then
z
1
(y, S(t), T) = 0 and z
2
(y, S(t), T) = (S(T) K)
+
and the option price
(x, y, S(t), t;
1

, 1) = z
2
(y, S(t), t) z
1
(y, S(t), t).
By the theory of stochastic optimal control [10], Barles and Soner state that the value func-
tions V
1
and V
2
are the unique solutions of the dynamic programming equation
min{V
t
+
1
2

2
S
2
V
SS
rSV
S
, V
y
+S(1 +)V
x
, V
y
S(1 )V
x
} = 0,
which leads to a dynamic programming equation for z
1
and z
2
, which are independent of
the variable x.
Supposing that the proportional transaction cost is equal to a

for some constant

and (x) is the solution to the following

(x) =
(x) + 1
2
_
x(x) x
, x = 0, (19a)
with the initial condition
(0) = 0. (19b)
Nonlinear Models in Option Pricing an Introduction 15
The analysis of this ODE (19) by Barles and Soner in [2] implies that
lim
x
(x)
x
= 1 and lim
x
(x) = 1. (20)
The property (20) encourages to treat the function () as the identity for large arguments
and therefore to simplify the calculations. In this case the volatility becomes

2
=
2
(1 +e
r(Tt)
a
2
S
2
V
SS
). (21)
The existence of a viscosity solution to (2) for European options with the volatility given
by (18) is proved by Barles and Soner in [2] and their numerical results indicate an eco-
nomically signicant price difference between the standard BlackScholes model and the
nonlinear model with transaction costs.
6.4 Risk Adjusted Pricing Methodology
In this model, proposed by Kratka in [22] and improved by Janda cka and ev covi c in [20],
the optimal time-lag t between the transactions is found to minimize the sum of the rate
of the transaction costs and the rate of the risk from an unprotected portfolio. That way the
portfolio is still well protected with the Risk Adjusted Pricing Methodology (RAPM) and
the modied volatility is now of the form

2
=
2
_
1 + 3
_
C
2
M
2
SV
SS
_1
3
_
, (22)
where M 0 is the transaction cost measure and C 0 the risk premium measure.
It is worth mentioning that these nonlinear transaction cost models that are described
above are all consistent with the linear model if the additional parameters for transaction
costs are equal to zero and vanish (Le, (), M).
Conclusion
In this chapter we provided a profound overview over nonlinear BlackScholes equations
for European and American options.
We introduced the reader to the nancial terminology and to BlackScholes equations
and presented several reasons for their nonlinearity and focused on the nonlinearity resulting
from a modied volatility function due to transaction costs. Here we focused on several
transaction cost models, including Leland model, Barles and Soners model, the identity
model and the Risk Adjusted Pricing Methodology.
Acknowledgements
The author acknowledges very fruitful discussions with Daniel ev covi c in the framework
of a bilateral GermanSlovakian DAAD project Fin-Diff-Fin: Finite differences for Finan-
cial derivative models.
16 Matthias Ehrhardt
Appendix
A Stochastics
In this chapter, we used several terms and concepts of probability theory and stochastics.
Thus, we recall some denitions (see e.g. [13, 27, 28] and the references therein).
A.1 Probability Space
Let be a sample space representing all possible scenarios (e.g. all possible paths for the
stock price over time). A subset of is an event and a sample point.
Denition A.1 Let be a nonempty set and F be a collection of subsets of . F is called
a -algebra (not related to the volatility ), if
i) F,
ii) whenever a set A belongs to F, its complement A
c
also belongs to F and
iii) whenever a sequence of sets A
n
, n N belongs to F, their union

n=1
A
n
also
belongs to F.
In our nancial scenario, F represents the space of events that are observable in the market
and therefore, all the information available until the time t can be regarded as a -algebra
F
t
. It is logical that F
t
F
s
for t < s, since the information that has been available t is
still available at s.
Denition A.2 Let be a nonempty set and F be a -algebra of subsets of . A probability
measure P is a function that assigns a number in [0, 1] to every set A F. The number is
called the probability of A and is written P(A). We require:
P() = 1 and
whenever a sequence of disjoint sets A
n
, n N belongs to F, then
P
_

_
n=1
A
n
_
=

n=1
P(A
n
).
The tripel (, F, P) is called a probability space.
A.2 Random Variable
Denition A.3 A real-valued function X on is called a random variable if the sets
{X x} := { : X() x} = X
1
(] , x])
are measurable for all x R. That is, {X x} F.
Nonlinear Models in Option Pricing an Introduction 17
A.3 Stochastic Process
Denition A.4 A (continuous) stochastic process X(t) = X(, t), t [0, [, is a family of
random variables X : [0, [R with t X(, t) continuous for all .
A.4 It Process
Denition A.5 An It process is a stochastic process of the form
dX = a(X, t)dt +b(X, t)dW,
which is equivalent to
X(t) = X(0) +
_
t
0
a(X, s)ds +
_
t
0
b(X, s)dW,
where X(0) is nonrandom, W(t) is a standard Wiener process, a() and b() are sufciently
regular functions and the integrals are It integrals.
A.5 Stopping Time
Denition A.6 A stopping time t is a random variable taking values in [0, ] and satisfy-
ing
{t s} F
s
s 0.
A.6 Brownian Motion
Denition A.7 A Brownian motion or Wiener process is a time-continuous stochastic pro-
cess W(t) with the properties:
W(0) = 0.
W(t) N(0, t) for all t 0. That is, for each t the random variable W(t) is nor-
mally distributed with mean E[W(t)] = 0 and variance Var[W(t)] = E[W
2
(t)] = t.
All increments W(t) := W(t +t) W(t) on non-overlapping time intervals are
independent. That is, W(t
2
) W(t
1
) and W(t
4
) W(t
3
) are independent for all
0 t
1
< t
2
t
3
< t
4
.
W(t) depends continuously on t.
18 Matthias Ehrhardt
A.7 Its Lemma
Theorem A.8 Consider a function V (S, t) : R [0, [ R with V C
2,1
(R [0, [)
and suppose that S(t) follows the It process
dS = a(S, t)dt +b(S, t)dW,
where W(t) is a standard Wiener process. Then V follows an It process with the same
Wiener process W(t):
dV = (aV
S
+
1
2
b
2
V
SS
+V
t
)dt +bV
S
dW, (23)
where a := a(S, t) and b := b(S, t).
If we consider a special case, where a(S, t) = S and b(S, t) = S, then S(t) follows
the Geometric Brownian motion, where W(t) is a standard Wiener process, and we have
dS = Sdt +SdW.
Then, Its Lemma yields
dV = (SV
S
+
1
2

2
S
2
V
SS
+V
t
)dt +SV
S
dW
=
_
1
2

2
S
2
V
SS
+V
t
_
dt +V
S
dS.
B Pricing Formulae
Theorem B.1 The solution to the linear BlackScholes equation (1) with the terminal and
boundary conditions (4), or the value of the European Call option, is given by
V (S, t) = Se
q(Tt)
N(d
1
) Ke
r(Tt)
N(d
2
), (24)
where
d
1
:=
ln
_
S
K
_
+ (r q +

2
2
)(T t)

T t
d
2
:=
ln
_
S
K
_
+ (r q

2
2
)(T t)

T t
and N(x) is the standard normal cumulative distribution function
N(x) =
1

2
_
x

y
2
2
dy, x R.
Respectively, the value of the European Put option is the solution to the linear BlackScholes
equation (1) with the terminal and boundary conditions (5) and is given by
V (S, t) = Se
q(Tt)
N(d
1
) Ke
r(Tt)
N(d
2
). (25)
Nonlinear Models in Option Pricing an Introduction 19
References
[1] M. Avellaneda and A. Pars, Dynamic hedging portfolios for derivative securities in
the presence of large transaction costs, Appl. Math. Finance 1 (1994), 165193.
[2] G. Barles and H.M. Soner, Option pricing with transaction costs and a nonlinear
BlackScholes equation, Finance Stoch. 2 (1998), 369397.
[3] F. Black and M. Scholes, The pricing of options and corporate liabilities, J. Polit.
Econ. 81 (1973), 637659.
[4] L. Bordag and A. Chmakova, Explicit solutions for a nonlinear model of nancial
derivatives, Int. J. Theor. Appl. Finance 10 (2007), 121.
[5] P. Boyle and T. Vorst, Option Replication in Discrete Time with Transaction Costs,
J. Finance 47 (1992), 271293.
[6] A. Chmakova, Symmetriereduktionen und explizite Lsungen fr ein nichtlineares
Modell eines Preisbildungsprozesses in illiquiden Mrkten, Dissertation, Technische
Universitt Cottbus, 2005 (in german).
[7] B. Dupire, Pricing with a Smile, Risk 7 (1994), 1820.
[8] M. Ehrhardt and R.E. Mickens, A fast, stable and accurate numerical method for the
BlackScholes equation of American options, Int. J. Theoret. Appl. Finance 11 (2008),
471501.
[9] G. Ferreyra, The Mathematics Behind the 1997 Nobel Prize in Economics, e-MATH,
Whats New in Mathematics, (electr. magazine of the Amer. Math. Soc.), 1997.
[10] W.H. Fleming and H.M. Soner, Controlled Markov Processes and Viscosity Solutions,
Springer, New York, 1993.
[11] R. Frey and A. Stremme, Market Volatility and Feedback Effects from Dynamic Hedg-
ing, Math. Finance 4 (1997), 351374.
[12] R. Frey and P. Patie, Risk Management for Derivatives in Illiquid Markets: A Simula-
tion Study, Springer, Berlin, 2002.
[13] M. Gnther and A. Jngel, Finanzderivate mit MATLAB, Vieweg, 2003 (in german).
[14] H. Han and X. Wu, A fast numerical method for the BlackScholes equation of Amer-
ican options, SIAM J. Numer. Anal. 41 (2003), 20812095.
[15] S. Heston, A ClosedForm Solution for Options with Stochastic Volatility with Appli-
cations to Bond and Currency Options, Rev. Fin. Studies 6 (1993), 327343.
20 Matthias Ehrhardt
[16] S. Hodges and A. Neuberger, Optimal replication of contingent claims under transac-
tion costs, Review of Futures Markets 8 (1989), 222239.
[17] T. Hoggard and A. E. Whalley and P. Wilmott, Hedging option portfolios in the pres-
ence of transaction costs, Adv. Fut. Opt. Res. 7 (1994), 2135.
[18] J. Hull and A. White, The Pricing of Options on Assets with Stochastic Volatilities,
J. Finance 42 (1987), 281300.
[19] J. Hull, Options, Futures and Other Derivatives, 5th Edition, Prentice Hall, 2002.
[20] M. Janda cka and D. ev covi c, On the risk-adjusted pricing-methodology-based valu-
ation of vanilla options and explanation of the volatility smile, J. Appl. Math. 3 (2005),
235258.
[21] Y. Kabanov and M. Safarian, On Lelands Strategy of Option Pricing with Transac-
tions Costs, Finance Stoch. 1 (1997), 239250.
[22] M. Kratka, No Mystery Behind the Smile, Risk 9 (1998), 6771.
[23] Y.-K. Kwok, Mathematical Models of Financial Derivatives, Springer Finance, 1998.
[24] H.E. Leland, Option pricing and replication with transactions costs, J. Finance 40
(1985), 12831301.
[25] R.C. Merton, Theory of rational option pricing, Bell J. Econ. Manag. Sci. 4 (1973),
141183.
[26] P. Schnbucher and P. Wilmott, The feedbackeffect of hedging in illiquid markets,
SIAM J. Appl.Math. 61 (2000), 232272.
[27] R. Seydel, Tools for Computational Finance, Second ed., Springer, 2004.
[28] S.E. Shreve, Stochastic Calculus for Finance II: ContinuousTime Models, Springer
Finance, New York, 2004.
[29] H.M. Soner, S.E. Shreve and J. Cvitani c, There is No Nontrivial Hedging Portfolio for
Option Pricing with Transaction Costs, Ann. Appl. Probab. 5 (1995), 327355.
[30] H.M. Soner and N. Touzi, Superreplication Under Gamma Constraints, SIAM
J. Contr. Optim. 39 (2001), 7396.
[31] D. Tavella and C. Randall, Pricing nancial instruments: The nite difference method,
John Wiley & Sons, 2000.
[32] P. Wilmott, J. Dewynne and S. Howison, Option Pricing: Mathematical Models and
Computation, Oxford, Financial Press, 1993.
Nonlinear Models in Option Pricing an Introduction 21
[33] P. Wilmott, S. Howison, and J. Dewynne, The Mathematics of Financial Derivatives,
A Student Introduction, Cambridge University Press, 2002.
[34] Y. Zhu, X. Wu and I. Chern, Derivative securities and difference methods, Springer
Finance, New York, 2004.