Anda di halaman 1dari 20

Goethe-University Frankfurt am Main

T ERM PAPER

Pricing Double Barrier Options with Finite Difference Methods


Seminar: Finance and MATLAB Summer Term 2011

submitted to: Professor Dr. Christian Schlag Chair of Derivatives and Financial Engineering Faculty of Economics and Business Administration Johann Wolfgang Goethe-University Frankfurt am Main

by: stud. rer. pol. Torsten Radtke Offenbacher Landstr.416 60599 Frankfurt am Main Student Number: 4455130 Field of Study: Money and Finance

Frankfurt am Main, 13.06.2011

Pricing Double Barrier Options with Finite Difference Methods

Contents
1 Introduction 2 Derivation of the Fundamental PDE 3 Finite Difference Methods 3.1 3.2 3.3 Explicit nite-difference method . . . . . . . . . . . . . . . . . . . . . . Fully implicit nite-difference method . . . . . . . . . . . . . . . . . . . Crank-Nicolson nite-difference method . . . . . . . . . . . . . . . . . . 1 1 3 5 6 7 7 10

4 Application to Double Barrier Options 5 Conclusion

Pricing Double Barrier Options with Finite Difference Methods

II

List of Figures
1 2 3 4 5 6 An example of a grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . Explicit scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Fully implicit scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . Crank Nicolson scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . Matlab Output: American Double Barrier . . . . . . . . . . . . . . . . . Matlab Outut: European Double Barrier . . . . . . . . . . . . . . . . . . 6 6 7 7 11 11

Pricing Double Barrier Options with Finite Difference Methods

1 Introduction
This term paper is intended to introduce and explain nite difference methods in the context of pricing double barrier options. Chapter 2 describes how to derive the fundamental partial differential equation (PDE) in the ground-breaking option pricing model proposed by Black and Scholes (1973)1 , which has remained very inuential up to now. The chapter introduces the theoretical framework of the main ideas in option pricing and states the PDE, which every contingent claim has to solve. A contingent claim is a nancial product, whose value is contingent on the evolution of a specied underlying. Since there do not exist analytical solutions for every contingent claim, numerical procedures can be applied to get an approximation. Chapter 3 describes one eld of numerical techniques, namely the nite difference methods, which were rst suggested by Schwartz (1977), and Brennan & Schwartz (1978)2 . After an explanation of how the three most common nite difference methods work in general, we apply them to double barrier options in chapter 4. The term paper will be completed by a conclusion.

2 Derivation of the Fundamental PDE


This chapter describes and explains how to derive the fundamental partial differential equation in arbitrage pricing theory3 . First of all, the assumptions of the underlying model will be presented and some key tools in option pricing will be explained to get the framework we need. In a second step one of the several ways to derive the famous equation will be illustrated. The economic assumptions behind Black-Scholes are as follows4 : (A1) Short Selling is allowed with full use of proceeds (A2) Markets are liquid (A3) Constant risk-free interest rates r (A4) Trading takes place continuously (A5) No transaction costs (A6) No arbitrage opportunities in the market (A7) The underlying S follows a certain lognormal process
1 2

[BS73] [BS78] 3 e.g.[Chr97, Chapter 4] 4 see [Chr97, Chapter 5.5]

Pricing Double Barrier Options with Finite Difference Methods

Although it can be shown that all of the above assumptions are unrealistic to a greater or lesser extent, the Black-Scholes-Merton (BSM) model is of profound importance to practitioners and in theory as well. As we know from (A7) the underlying follows a geometric Brownian motion dSt = St dt + St dWt (1)

where is the drift and is the volatility of the underlying. The rst term can be seen as the expected return in an innitesimal small time step and the second term is the instantaneous standard deviation of the return. Both are parameters and thus do not change through the maturity T of the option. Since we know that the value of the option has to be a function of time and of the underlying asset V (t, S), we can apply Itos lemma5 to get dVt = V 1 2V V dt + dt + 2 S 2 2 dt t S 2 S (2)

Furthermore there must be a certain correlation between the underlying and the option, which we can exploit to construct a special portfolio , consisting of an option in long position and a short position in the underlying in some quantity . = V (S, t) S Thus the portfolio changes by d = V 1 2V V dt + dS + 2 S 2 2 dt dS t S 2 S (4) (3)

Because our preferential aim is to nd an applicable equation for the option price V , we should rst get rid of the randomness in our portfolio. This can be done by choosing the right delta in our portfolio since the whole randomness from the portfolio comes from the change in S. Thus if we choose =
V S

the random terms cancel out and the change

in the portfolio becomes deterministic. This approach is generally called delta hedging and it is a key building block of the asset pricing theory. The idea behind that concept goes back to Higgins (1902) & Nelson (1904), who introduced static hedging. In case of our portfolio, a dynamic delta hedge is needed, because the quantity =
V S

changes

for every single time step and it must be rebalanced continually. The dynamic hedging approach was rst described by Thorp & Kassouf (1967)6 . Applying this principle we get ( ) V 1 2 2 2V d = + S dt (5) t 2 S 2
5 6

e.g.[Bjo09, Chapter 4] [TK67]

Pricing Double Barrier Options with Finite Difference Methods

Since this change is completely riskless and (A6) holds, we can conclude that the return of our portfolio is the risk free rate and thus d = rdt (6)

This follows by the no arbitrage principle, which is the second important building block we need for the derivation. Now we just have to substitute (3) into (6) and then (7) and after rearranging we get V V 2V + rS + 2 S 2 2 rV = 0 t S S (7)

This is the Black-Scholes-Merton equation and it was published in 1973. The crucial fact is that every function of a contingent claim which depends on the underlying S has to solve this equation. One reason why this equation has become so famous is that the rational option price in this model does not depend on any drift term of the underlying. Therefore the model is independent of any assumption on the preferences of investors. This is the big difference to older option pricing models introduced as for instance by Sprenkle (1964), Boness (1964) or Samuelson (1965). You can get closed form solutions for the BSM equation analytically to value simple contingent claims such as European call options by exploiting the boundary conditions of these contracts. But we will now focus on numerical solutions for this equation to get a tool for pricing exotic options with more complex boundary conditions.

3 Finite Difference Methods


The idea behind all nite difference methods is the approximation of every partial derivative of the PDE by a difference quotient. To obtain these difference quotients, we can make use of Taylors theorem7 , which states that under suitable continuity and differentiability conditions a function f (x) may be represented as 1 1 f (x + h) = f (x) + hf (x) + h2 f (x) + h3 f (x) + ... 2 6 (8)

where h is the size of the change in x. We can now get the rst-order derivative by neglecting the terms of h2 and higher f (x) =
7

f (x + h) f (x) + O(h) h

(9)

e.g.[BB, chapter 5]

Pricing Double Barrier Options with Finite Difference Methods

O(h) describes the so called truncation error. As the rst-order derivative is just dened as a limit of the above difference quotient, the truncation error vanishes if h converges to zero. Thus the quotient is an approximation of a rst-order derivative. It is called a forward difference. To derive the backward difference, we just can use similar reasoning and obtain f (x) = f (x) f (x h) + O(h) h (10)

In this case we are also facing a truncation error of order O(h). The third approximation we will use for our purposes can be obtained by subtracting (11) from (9) and rearranging: f (x) = f (x + h) f (x h) + O(h2 ) 2h (11)

This is the central or symmetric difference and it has the advantage that the truncation error is only of order O(h2 ). Hence this approximation is better than the forward and backward difference in terms of accuracy. That comes from the fortunate cancelation of terms, due to the symmetry about x in the denition of the difference. However, in some cases it might be useful to choose one of the one-sided differences as an efcient numerical scheme, depending on the type of boundary conditions. In order to discretize the BSM PDE afterwards, we have to nd approximations for second-order derivatives, too. One way of calculating it is to add [9] and [11] and to rearrange, which yields f (x) = f (x + h) 2f (x) + f (x h) + O(h2 ) h2 (12)

As we can see, the truncation error in this case is of order O(h2 ). Now we have all necessary discretization tools we need for the following nite difference schemes. In order to apply the approximations of the derivatives to the BSM PDE we have to set up a discrete grid of points of the form (it, jS) with t and S as the size of the discretization steps, where 0 i I and 0 j J. Hence It equals the maturity date of the option and JS is the upper bound whose value depends on the boundary conditions of the contract. Every node stands for one option value V (t, S) = Vi,j = V (it, jS). Furthermore it is our aim to calculate the option values for every node in the grid by using nite difference methods. The grid points we are nally interested in are on the y-axes because we want to nd a rational option price for t = 0. An example of an appropriate grid is shown in Figure 1. We will now introduce three diverse nite-difference schemes, namely the explicit, the fully implicit and the Crank-Nicolson method. They only differ

Pricing Double Barrier Options with Finite Difference Methods

in depending on the algorithms, which calculate the grid points. Basically, they are just using different time derivatives. Before we apply the above tools, we need to transform the BSM PDE in order to simplify the numerical analysis afterwards. As shown by Brennan & Schwartz (1978) and Hull & White (1990)8 it is more efcient to use lnS rather than S as the underlying variable. This is because the changes in lnS in a time interval t are independent of S and t. According to Brennan & Schwartz we have to dene Z lnS and W (t, Z) V (t, S) to obtain ( ) 1 2 2W 1 2 W W + r + rW = 0 2 2 Z 2 Z t

(13)

In terms of the grid, we have to consider that we need Z = lnS on the y-axis and we have to dene the size of the increment Z. Furthermore the option values on the grid points are dened as Wi,j = W (t, Z) = W (it, jZ). In order to evaluate the three schemes we need to dene some criteria. The rst criterion is consistency: A consistent nite difference scheme has a decreasing difference approximation error as the increment size of the underlying and the time step size approach zero. The second one is stability: A numerical system is stable of the errors coming from any source do not grow with time. The most important criterion is convergence: It means that the solution to a FDE converges to the true solution as the increment size is reduced. Lax equivalence theorem9 states that for a consistent nite difference method for a well-posed linear initial value problem, the method is convergent if and only if it is stable.

3.1 Explicit nite-difference method


We will rst derive the explicit scheme to start with the simplest one. For the discretization of the transformed BSM PDE we use the forward difference for the time derivative and the already-mentioned central differences for (13) becomes 1 Wi,j+1 Wi,j1 Wi,j Wi1,j + (r 2 ) t 2 2Z 1 2 Wi,j+1 + Wi,j1 2Wi,j + rWi,j = 0 2 Z 2 After rearranging we get
8 9

W Z

and

2W . Z 2

Thus, the approximation of

(14) (15)

[HW90] [Duf06, chapter8]

Figure 1: An example of a uniform grid points in S-t space [Source: Authors design]

Figure 2: The explicit method [Source: Authors design]

Pricing Double Barrier Options with Finite Difference Methods

with ex

Wi,j = ex Wi+1,j1 + ex Wi+1,j + ex Wi+1,j+1 [ ] [ [ ] 2 t(r ) 1 t 2 1 t 2 1 t 2 2 = 1+rt 2Z 2 2Z ex = 1+rt 1 Z 2 ex = 1+rt 2Z 2 +

t(r ) 2 2Z

(16) ] 2

We want to calculate the left hand side (LHS) with the help of the option prices on the RHS. Figure 2 illustrates the procedure in terms of grid points. It is possible to solve the equations explicitly as the name already suggests. As Brennan & Schwartz stated, the scheme is stable when the weights ex , ex and ex are nonnegative. This gives us the stability condition t
( 1
2 Z

)2 .

This relationship makes the scheme slower, because you

need to adjust the size of the time step when you decrease Z. Moreover, if the condition is fullled the Laxs equivalence theorem states, that the nite different scheme is stable and convergent. In terms of accuracy and consistency, we know that the time derivative has an error of O(t) and the central difference of the stock of O(Z 2 ). Therefore the method is accurate to (t, Z 2 ). The advantage of this method lies in the simplicity of the solution, which is the reason why it is easy to program. On the other hand we are facing restrictions on the time step, which can lead to slower convergence.

3.2 Fully implicit nite-difference method


The derivation of the fully implicit scheme works similar to the above approach. We just use a backward difference for the time derivative and obtain after the appropriate rearrangements im Wi,j1 + im Wi,j + im Wi,j+1 = Wi+1,j (1 rt) with: im =
t 2 2Z 2

(17)
t(r ) 2 2Z
2

t(r ) 2 2Z

and im = 1 +

2 t Z 2

and im =

2 +2 2Z

Figure 3 shows the relationship between option values in this method. As we have to calculate the option values on the LHS from the option value on the RHS, the solution is much more complex than the explicit one, but you can get solutions faster, because there are no restrictions on the size of time step. Due to Laxs theorem, you can see that under the above condition, the fully implicit scheme is also convergent. Due to the application of the backward difference and the central difference this scheme is also accurate to O(t, Z 2 ), as in the explicit scheme.

Figure 3: The implicit method [Source: Authors design]

Figure 4: The Crank-Nicolson method [Source: Authors design]

Pricing Double Barrier Options with Finite Difference Methods

3.3 Crank-Nicolson nite-difference method


The Crank-Nicolson scheme is a combination of the two above methods. As shown in gure 4 it uses the 3 option values on the RHS to calculate the three option values on the LHS implicitly. To obtain the following equations, we have to go one half imaginary step forward in time, which can be achieved by averaging the two above schemes. To obtain the Crank-Nicolson scheme we do not stick to the log transformed version of the BSM PDE as it was proposed by Brennan & Schwartz for the explicit and fully implicit scheme. Thus we discretize the untransformed BSM PDE by averaging the two methods: Vi+1,j Vi,j rjS + [Vi,j+1 Vi,j1 + Vi+1,j+1 Vi+1,j1 ] t 4S 2 j 2 S 2 + [Vi,j+1 + Vi,j1 2Vi,j + Vi+1,j+1 + Vi+1,j1 2Vi+1,j ] 4S 2 r [Vi+1,j + Vi,j ] = 0 2 After rearranging we obtain cn Vi,j1 + (cn + 1)Vi,j + cn Vi,j+1 = cn Vi+1,j1 (cn 1)Vi+1,j cn Vi+1,j+1 with: cn (j) = t ( 2 j 2 rj) and cn (j) = 4
t ( 2 j 2 2

(18) (19) (20)

(21) (22)

+ r) and cn (j) = t ( 2 j 2 + rj). 4

The LHS has to be solved implicitly, too. This is why the Crank-Nicolson method belongs to the implicit schemes. Although the set of equations is a bit more difcult to solve, this scheme is the most efcient one and it can be shown that it has an accuracy of (t2 , S 2 ). As you can see the coefcients are not constant anymore, but they depend on the price of the underlying S = jS.

4 Application to Double Barrier Options


In order to apply the described numerical methods to double barrier options, we have to become familiar with those contracts. Barrier options have path-dependent payoffs. This is because the payoff is dependent upon whether the underlying reaches certain specied barrier levels. There are many types of these contracts, but the two most common ones are knock-out and knock-in barrier options. While the knock-out option becomes worthless if the underlying asset crosses a specied barrier level, the knock-in option will become alive only if the underlying crosses a certain barrier. Furthermore you can divide between

Pricing Double Barrier Options with Finite Difference Methods

up barriers and down barriers. But we will focus on double barrier options, which typically have one up barrier U and one down barrier L. This means for a double barrier knock-out option the value of the option would become worthless if the underlying asset drops below the down barrier or rises above the up barrier. Thus the upper boundary is dened by the up barrier and the lower boundary is dened by the down barrier of the contract. An investor would choose such a contract if he believes changes in the underlying asset price would be within a narrow range. You can choose between two exercise types: The American and the European style. The rst one gives the holder of the option the right to exercise the option at any time up to the expiration date, while the European option can be exercised only on the maturity date. We will focus on the two common option types put options and call options. These characteristics determine the payoff of the option. In case of a call option the holder has the right to buy the underlying asset for a certain price by a certain date. The holder of a put option has the right to sell the underlying asset for a certain price by a certain date. The specied price in the contract is known as the strike price K. Consider now a double barrier call option with two knock out barriers. We can write the terminal condition as V (S, T ) = VI,j = (Sj K)+ for L < S < U . The general solution system can be written as a1 1 + b1 c1 0 ... ... ... 0 Vi,0 . . Vi,1 0 a2 1 + b2 c2 0 ... ... . . . . . Vi,2 = . 0 a3 1 b3 c3 0 ... . . . . ... ... ... ... ... . . . . 0 . . ... ... ... ... 0 aJ1 1 + bJ1 cJ1 Vi,J a 1 b1 c1 0 ... ... ... 0 V i+1,0 1 . . Vi+1,1 0 a2 1 b2 c2 0 ... ... . . . . Vi+1,2 . . 0 a3 1 b3 c3 0 ... . . . . ... ... ... ... ... . . . . . 0 . ... ... ... ... Vi+1,J 0 aJ1 1 bJ1 cJ1 We have to solve these equations10 for every time step i = 1, 2.., I. We can write this as AV (i) = BV (i + 1).The values of the elements in the matrices A and B depend on the nite difference schemes. In case of the Crank-Nicolson scheme aj = cn (j),
10

see[DM04, pp130]

Pricing Double Barrier Options with Finite Difference Methods

bj = cn (j) and cj = cn (j). In the above notation the RHS is completely deterministic, since it has been calculated in the previous time step and the initial values are given by the terminal condition. Since we have (J 1) equations for (J + 1) unknowns, we need two additional equations two solve the system. Now we need the boundary conditions. Since we know that the option is worthless when it reaches the upper or lower boundary, we will just calculate the prices within this range. Thus we plug in the upper boundary V (U, t) = V (JS, it) = Vi,J = 0 and lower boundary V (L, t) = V (0, it) = Vi,0 = 0. Notice that Sj becomes Sj = L + jS in this notation. When we now leave out a1 Vi,J and cJ1 Vi,0 on both sides, A and B become symmetric matrices of order (J 1). You can now invert A and obtain V (i) = A1 BV (i + 1). In case of the explicit scheme A is an identity matrix, in case of the implicit scheme B is an identity matrix and the respective elements equal the above explained coefcients of the particular method. The inversion of an identity matrix just gives the identity matrix and this is why the explicit scheme is easy to solve. For the implicit schemes it is more costly to calculate the unknown option prices. But a calculation intensive inversion can be avoided by using several techniques, for example the L-U decomposition11 . It is now possible to calculate every grid point going backward in time up to t = 0. You have to nd the specic grid point, where the option price for the current price of the underlying S0 is located. If this point is not on the grid, you have to interpolate between the two closest grid points. The above procedure generates the price of a European double barrier call option. To implement the American style exercise you just have to check, whether the price of the option Vi,j or the payoff (Sj K)+ is greater on every grid point right after it has been calculated and thus change it if the payoff is bigger, because a rational investor would exercise the option in this scenario. Figure 5 and gure 6 show the surface of the option price matrix in case of the European and the American exercise type simulated by matlab. You can nd the matlab code and relevant explanations in the appendix. The prices have been calculated with the assumption that it is monitored continuously whether the price of the underlying hits the barriers or not. But double barrier option contracts can also have the feature of discrete monitoring dates. This is a big problem for the nite difference methods we discussed so far because the barriers appear inside the solution. While we could calculate the solutions under the upper and above the lower barrier in the continuously monitoring case, we
11

e.g.[Wil98]

Pricing Double Barrier Options with Finite Difference Methods

10

need both of them within the solution system for discrete monitoring. It is not possible to solve the PDE at these points because of their discontinuity property. At these points the options delta
V S

is discontinuous which leads to spurious oscillations in the option

price near the barriers. This problem was for instance discussed by Tagliani (2009)12 . He proposed three very time consuming schemes with very small time steps. He introduced a variant of the Crank-Nicolson scheme, in which the explicit and implicit scheme are not averaged but weighted in another way. He also explained a semiimplicit scheme and another implicit scheme with lower accuracy than the Crank-Nicolson variant. They all can guarantee the absence of oscillations close to the monitoring dates, but they need a very high number of discretization nodes.

5 Conclusion
We have shown that nite difference methods can be powerful tools for pricing contingent claims if no analytical solution is available. The approach to discretize the PDE, which has to hold for every contingent claim is as simple as efcient. Especially in case of American double barrier options, the nite difference approach delivers appropriate results in contrast to the Monte Carlo simulation for instance. On the other hand you can hardly use the nite difference approach for strongly path-dependent options such as Asian or lookback options due to the fact that the value of the payoff depends directly on the path of the underlying. Thus you always have to mind the type of contingent claim you are facing. When we compare the three mentioned nite difference methods the explicit scheme is just equivalent to a trinomial lattice approach as shown by Brennan & Schwartz (1989). The implicit method in comparison does not have a restriction on the size of the time step and can converge faster. Furthermore the Crank Nicolson method is very popular because of its stability and the high order of accuracy. Although numerous articles in the nance literature use this scheme, it has problems in certain situations as already mentioned in the prior chapter. You can nd more robust alternatives in many cases where the boundary conditions are too complex. In order to choose a certain nite difference scheme, you always have to consider the already mentioned evaluation criteria: stability, convergency and consistency.
12

[Tag09]

Figure 5: The surface of an option price matrix for an American double barrier option [Source: Authors design]

Figure 6: The surface of an option price matrix for a European double barrier option [Source: Authors design]

Pricing Double Barrier Options with Finite Difference Methods

12

Appendix
The following code uses the Crank-Nicolson method, because it is the most efcient one. You can easily implement the explicit or the implicit method by just choosing other other for the A and B coefcient matrices. If you want to implement the coefcients for the log transformed PDE, you have to be careful with the boundary conditions. The terminal condition for a call becomes VI,j = (eZ K)+ for example.

Matlab Code
% CrankNicolsonDoubleBarrier.m % This program suits the purpose of pricing Double Barrier Options using the % Crank-Nicolson scheme to solve the Black-Scholes PDE. With some minor % adjustments, it is also possible, to evaluate plain vanilla options % (note that in this case not only the input parameters have to be changed, % but also the terminal and boundary conditions below have to be adjusted). % The input parameters can be chosen more or less arbitrarily, and the output % is a price matrix and a possibly interpolated price, corresponding to the % initial stock value S0.

clear all; clc;

% Input parameters S0=100; K=100; U=150; L=25; s=0.2; r=0.04; T=1; EX=0; OP=1; I=50; J=50; % Initial stock value % Strike price % Upper boundary (for plain vanilla options set U=4*S0) % Lower boundary (for plain vanilla options set L=0) % Annualized volatility % Constant risk free rate % Maturity measured in years % Exercise type, american=1, european=else % Option type, call=1, put=-1 % Number of time steps % Number of steps in underlying value

% Calculation of necessary constants dT=T/I; dS=(U-L)/J; % Time step size % Size of steps in underlying value

Svec=(L:dS:U); % underlying value vector from lower to upper boundary jvec=Svec/dS; ivec=0:I; % underlying value grid vector % time step grid vector

a=-0.25*dT*(s2*(jvec.2)-r*jvec); %Coefficients from difference quotients b=dT*0.5*(s2*(jvec.2)+r); c=-0.25*dT*(s2*(jvec.2)+r*jvec);

% Calculations of the Coefficient Matrices A and B A=diag(a(3:J),-1)+diag(1+b(2:J))+diag(c(2:J-1),1); B=-diag(a(3:J),-1)-diag(-1+b(2:J))-diag(c(2:J-1),1);

V=zeros(J+1,I+1);

% Matrix for option price

% Implementing the boundary conditions V(:,I+1)=max((OP*(Svec-K)).*(Svec<U).*(Svec>L),0); % Terminal condition V(1,:)=0; V(J+1,:)=0; % Lower boundary % Upper boundary

Pricing Double Barrier Options with Finite Difference Methods

13

% Algorithm for Crank Nicolson scheme C=A\B; % Inversion of matrix A multiplied by B before the loop

for i= I : -1 : 1 V(2:end-1,i)=C*V(2:end-1,i+1); if EX==1 % If exercise type is american,

V(2:end-1,i)=max(V(2:end-1,i),V(2:end-1,J+1)); % comparison of time value and immediate exercise end end

%Output price=interp1(Svec,V(:,1),S0) surf(V) xlabel(time) ylabel(S) zlabel(option value) % If S0 is not on the grid, linear interpolation for the price w.r.t S0 % Displays the price matrix

Pricing Double Barrier Options with Finite Difference Methods

14

References
[BB] Paolo. Brandimarte and P. Brandimarte. Numerical methods in nance and economics. Wiley Online Library. [BGK97] M. Broadie, P. Glasserman, and S. Kou. A continuity correction for discrete barrier options. Mathematical Finance, 7(4):325349, 1997. [Bjo09] T. Bjoerk. Arbitrage theory in continuous time. Oxford University Press, USA, 2009. [BS73] F. Black and M. Scholes. The pricing of options and corporate liabilities. The journal of political economy, 81(3):637654, 1973. [BS78] M.J. Brennan and E.S. Schwartz. Finite difference methods and jump processes arising in the pricing of contingent claims: A synthesis. Journal of Financial and Quantitative Analysis, 13(03):461474, 1978. [Chr97] N. Chriss. Black-Scholes and beyond: option pricing models. McGraw-hill, 1997. [CON10] R. CONT. Encyclopaedia of quantitative nance (4 volume-set). 2010. [DM04] H.P. Deutsch and MyiLibrary. Derivatives and Internal Models. Palgrave Macmillan, 2004. [Duf06] D.J. Duffy. Finite difference methods in nancial engineering: a partial differential equation approach, volume 312. Wiley, 2006. [GC06] M.B. Giles and R. Carter. Convergence analysis of crank-nicolson and rannacher time-marching. Journal of Computational Finance, 9(4):89, 2006. [GJ03]
R M. Guenther and A. Juengel. Finanzderivate mit MATLAB. Vieweg+ Teub-

ner, 2003. [Hul09] J.C. Hull. Optionen, Futures und andere Derivate. Pearson Education, 2009.

[HW90] John Hull and Alan White. Valuing derivative securities using the explicit nite difference method. The Journal of Financial and Quantitative Analysis, 25(1):pp. 87100, 1990.

Pricing Double Barrier Options with Finite Difference Methods

15

[Tag09]

A. Tagliani. Discrete monitored barrier options by nite difference schemes. In Tenth Workshop in Quantitative Finance, Milano, 2009.

[TK67]

E.O. Thorp and S.T. Kassouf. Beat the market: a scientic stock market system. Random House, 1967.

[Top05]

J. Topper. Financial engineering with nite elements. John Wiley & Sons Inc, 2005.

[Wil98]

P. Wilmott. Derivatives. 1998.

Anda mungkin juga menyukai