Pricing Discrete Barrier Options

Transcription

1 Pricing Discrete Barrier Options Barrier options whose barrier is monitored only at discrete times are called discrete barrier options. They are more common than the continuously monitored versions. The main difficulty with pricing discrete barrier options lies in matching the monitored times. Here is why. Suppose each period has a duration of t and the l > 1 monitored times are t 0 = 0, t 1, t 2,..., t l = T. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 613

2 Pricing Discrete Barrier Options (continued) It is extremely unlikely that all monitored times coincide with the end of a period on the tree, meaning t divides t i for all i. The binomial-trinomial tree can handle discrete options with ease, however. We simply build a binomial-trinomial tree from time 0 to time t 1, followed by one from time t 1 to time t 2, and so on until time t l. See p c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 614

3 t 0 t 1 t 1 t 1 t 2 t 1 2σ t 1 { 2σ t 2 } c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 615

4 Pricing Discrete Barrier Options (concluded) This procedure works even if each t i is associated with a distinct barrier or if each window [ t i, t i+1 ) has its own continuously monitored barrier or double barriers. If the ith binomial-trinomial tree has n i size of the whole tree is ( l ) 2 O n i. i=1 periods, the c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 616

5 Options on a Stock That Pays Known Dividends Many ad hoc assumptions have been postulated for option pricing with known dividends. a 1. The one we saw earlier models the stock price minus the present value of the anticipated dividends as following geometric Brownian motion. 2. One can also model the stock price plus the forward values of the dividends as following geometric Brownian motion. a Frishling (2002). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 617

6 Options on a Stock That Pays Known Dividends (continued) The most realistic model assumes the stock price decreases by the amount of the dividend paid at the ex-dividend date. The stock price follows geometric Brownian motion between adjacent ex-dividend dates. But this model results in binomial trees that grow exponentially. The binomial-trinomial tree can often avoid the exponential explosion for the known-dividends case. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 618

7 Options on a Stock That Pays Known Dividends (continued) Suppose that the known dividend is D dollars and the ex-dividend date is at time t. So there are m t/ t periods between time 0 and the ex-dividend date. To avoid negative stock prices, we need to make sure the lowest stock price at time t is at least D, i.e., Se (t/ t)σ t D. Equivalently, t [ tσ ln(s/d) ] 2. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 619

8 Options on a Stock That Pays Known Dividends (continued) Build a binomial tree from time 0 to time t as before. Subtract D from all the stock prices on the tree at time t to represent the price drop on the ex-dividend date. Assume the top node s price equals S. As usual, its two successor nodes will have prices S u and S u 1. The remaining nodes successor nodes will have prices same as the binomial tree. S u 3, S u 5, S u 7,..., c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 620

9 Options on a Stock That Pays Known Dividends (concluded) For each node at time t below the top node, we build the trinomial connection. Note that the binomial-trinomial structure remains valid in the special case when t = t on p Hence the construction can be completed. From time t + t onward, the standard binomial tree will be used until the maturity date or the next ex-dividend date when the procedure can be repeated. The resulting tree is called the stair tree. a a Dai (R , D ) and Lyuu (2004). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 621

10 Multivariate Contingent Claims They depend on two or more underlying assets. The basket call on m assets has the terminal payoff ( m ) max α i S i (τ) X, 0, i=1 where α i is the percentage of asset i. Basket options are essentially options on a portfolio of stocks or index options. Option on the best of two risky assets and cash has a terminal payoff of max(s 1 (τ), S 2 (τ), X). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 622

11 Correlated Trinomial Model a Two risky assets S 1 and S 2 follow ds i /S i = r dt + σ i dw i in a risk-neutral economy, i = 1, 2. Let M i e r t, V i M 2 i (e σ2 i t 1). S i M i is the mean of S i at time t. S 2 i V i the variance of S i at time t. a Boyle, Evnine, and Gibbs (1989). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 623

12 Correlated Trinomial Model (continued) The value of S 1 S 2 at time t has a joint lognormal distribution with mean S 1 S 2 M 1 M 2 e ρσ 1σ 2 t, where ρ is the correlation between dw 1 and dw 2. Next match the 1st and 2nd moments of the approximating discrete distribution to those of the continuous counterpart. At time t from now, there are five distinct outcomes. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 624

13 Correlated Trinomial Model (continued) The five-point probability distribution of the asset prices is (as usual, we impose u i d i = 1) Probability Asset 1 Asset 2 p 1 S 1 u 1 S 2 u 2 p 2 S 1 u 1 S 2 d 2 p 3 S 1 d 1 S 2 d 2 p 4 S 1 d 1 S 2 u 2 p 5 S 1 S 2 c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 625

14 Correlated Trinomial Model (continued) The probabilities must sum to one, and the means must be matched: 1 = p 1 + p 2 + p 3 + p 4 + p 5, S 1 M 1 = (p 1 + p 2 ) S 1 u 1 + p 5 S 1 + (p 3 + p 4 ) S 1 d 1, S 2 M 2 = (p 1 + p 4 ) S 2 u 2 + p 5 S 2 + (p 2 + p 3 ) S 2 d 2. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 626

15 Correlated Trinomial Model (concluded) Let R M 1 M 2 e ρσ 1σ 2 t. Match the variances and covariance: S 2 1 V 1 = (p 1 + p 2 )((S 1 u 1 ) 2 (S 1 M 1 ) 2 ) + p 5 (S 2 1 (S 1M 1 ) 2 ) +(p 3 + p 4 )((S 1 d 1 ) 2 (S 1 M 1 ) 2 ), S 2 2 V 2 = (p 1 + p 4 )((S 2 u 2 ) 2 (S 2 M 2 ) 2 ) + p 5 (S 2 2 (S 2M 2 ) 2 ) +(p 2 + p 3 )((S 2 d 2 ) 2 (S 2 M 2 ) 2 ), S 1 S 2 R = (p 1 u 1 u 2 + p 2 u 1 d 2 + p 3 d 1 d 2 + p 4 d 1 u 2 + p 5 ) S 1 S 2. The solutions are complex (see text). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 627

16 Correlated Trinomial Model Simplified a Let µ i r σ2 i /2 and u i e λσ i t for i = 1, 2. The following simpler scheme is good enough: p 1 = p 2 = p 3 = p 4 = p 5 = 1 1 λ 2. [ ( 1 1 t µ 4 λ µ ) 2 + ρ ] λ σ 1 σ 2 λ 2, [ ( 1 1 t µ 4 λ µ ) 2 ρ ] λ σ 1 σ 2 λ 2, [ ( 1 1 t 4 λ 2 + µ 1 µ ) 2 + ρ ] λ σ 1 σ 2 λ 2, [ ( 1 1 t 4 λ 2 + µ 1 + µ ) 2 ρ ] λ σ 1 σ 2 λ 2, It cannot price 2-asset 2-barrier options accurately. b a Madan, Milne, and Shefrin (1989). b See Chang, Hsu, and Lyuu (2006) for a solution. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 628

17 Extrapolation It is a method to speed up numerical convergence. Say f(n) converges to an unknown limit f at rate of 1/n: f(n) = f + c ( ) 1 n + o. (70) n Assume c is an unknown constant independent of n. Convergence is basically monotonic and smooth. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 629

18 Extrapolation (concluded) From two approximations f(n 1 ) and f(n 2 ) and by ignoring the smaller terms, f(n 1 ) = f + c n 1, f(n 2 ) = f + c n 2. A better approximation to the desired f is f = n 1f(n 1 ) n 2 f(n 2 ) n 1 n 2. (71) This estimate should converge faster than 1/n. The Richardson extrapolation uses n 2 = 2n 1. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 630

19 Improving BOPM with Extrapolation Consider standard European options. Denote the option value under BOPM using n time periods by f(n). It is known that BOPM convergences at the rate of 1/n, consistent with Eq. (70) on p But the plots on p. 253 (redrawn on next page) demonstrate that convergence to the true option value oscillates with n. Extrapolation is inapplicable at this stage. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 631

20 Call value n Call value n c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 632

21 Improving BOPM with Extrapolation (concluded) Take the at-the-money option in the left plot on p The sequence with odd n turns out to be monotonic and smooth (see the left plot on p. 634). a Apply extrapolation (71) on p. 630 with n 2 = n 1 + 2, where n 1 is odd. Result is shown in the right plot on p The convergence rate is amazing. See Exercise of the text (p. 111) for ideas in the general case. a This can be proved; see Chang and Palmer (2007). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 633

22 Call value n Call value n c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 634

23 Numerical Methods c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 635

24 All science is dominated by the idea of approximation. Bertrand Russell c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 636

25 Finite-Difference Methods Place a grid of points on the space over which the desired function takes value. Then approximate the function value at each of these points (p. 638). Solve the equation numerically by introducing difference equations in place of derivatives. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 637

26 c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 638

27 Example: Poisson s Equation It is 2 θ/ x θ/ y 2 = ρ(x, y). Replace second derivatives with finite differences through central difference. Introduce evenly spaced grid points with distance of x along the x axis and y along the y axis. The finite difference form is ρ(x i, y j ) = θ(x i+1, y j ) 2θ(x i, y j ) + θ(x i 1, y j ) ( x) 2 + θ(x i, y j+1 ) 2θ(x i, y j ) + θ(x i, y j 1 ) ( y) 2. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 639

28 Example: Poisson s Equation (concluded) In the above, x x i x i 1 and y y j y j 1 for i, j = 1, 2,.... When the grid points are evenly spaced in both axes so that x = y = h, the difference equation becomes h 2 ρ(x i, y j ) = θ(x i+1, y j ) + θ(x i 1, y j ) +θ(x i, y j+1 ) + θ(x i, y j 1 ) 4θ(x i, y j ). Given boundary values, we can solve for the x i s and the y j s within the square [ ±L, ±L ]. From now on, θ i,j will denote the finite-difference approximation to the exact θ(x i, y j ). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 640

29 Explicit Methods Consider the diffusion equation D( 2 θ/ x 2 ) ( θ/ t) = 0. Use evenly spaced grid points (x i, t j ) with distances x and t, where x x i+1 x i and t t j+1 t j. Employ central difference for the second derivative and forward difference for the time derivative to obtain θ(x, t) t = θ(x, t j+1) θ(x, t j ) +, (72) t=tj t 2 θ(x, t) x 2 = θ(x i+1, t) 2θ(x i, t) + θ(x i 1, t) x=xi ( x) 2 +. (73) c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 641

30 Explicit Methods (continued) Next, assemble Eqs. (72) and (73) into a single equation at (x i, t j ). But we need to decide how to evaluate x in the first equation and t in the second. Since central difference around x i is used in Eq. (73), we might as well use x i for x in Eq. (72). Two choices are possible for t in Eq. (73). The first choice uses t = t j finite-difference equation, to yield the following θ i,j+1 θ i,j t = D θ i+1,j 2θ i,j + θ i 1,j ( x) 2. (74) c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 642

31 Explicit Methods (continued) The stencil of grid points involves four values, θ i,j+1, θ i,j, θ i+1,j, and θ i 1,j. Rearrange Eq. (74) on p. 642 as θ i,j+1 = D t ( x) 2 θ i+1,j + ( 1 2D t ( x) 2 ) θ i,j + D t ( x) 2 θ i 1,j. We can calculate θ i,j+1 from θ i,j, θ i+1,j, θ i 1,j, at the previous time t j (see exhibit (a) on next page). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 643

32 Stencils c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 644

33 Explicit Methods (concluded) Starting from the initial conditions at t 0, that is, θ i,0 = θ(x i, t 0 ), i = 1, 2,..., we calculate θ i,1, i = 1, 2,.... And then θ i,2, i = 1, 2,.... And so on. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 645

34 Stability The explicit method is numerically unstable unless t ( x) 2 /(2D). A numerical method is unstable if the solution is highly sensitive to changes in initial conditions. The stability condition may lead to high running times and memory requirements. For instance, halving x would imply quadrupling ( t) 1, resulting in a running time eight times as much. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 646

35 Explicit Method and Trinomial Tree Rearrange Eq. (74) on p. 642 as θ i,j+1 = D t ( x) 2 θ i+1,j + ( 1 2D t ( x) 2 ) θ i,j + D t ( x) 2 θ i 1,j. When the stability condition is satisfied, the three coefficients for θ i+1,j, θ i,j, and θ i 1,j all lie between zero and one and sum to one. They can be interpreted as probabilities. So the finite-difference equation becomes identical to backward induction on trinomial trees! The freedom in choosing x corresponds to similar freedom in the construction of trinomial trees. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 647

36 Implicit Methods Suppose we use t = t j+1 in Eq. (73) on p. 641 instead. The finite-difference equation becomes θ i,j+1 θ i,j t = D θ i+1,j+1 2θ i,j+1 + θ i 1,j+1 ( x) 2. (75) The stencil involves θ i,j, θ i,j+1, θ i+1,j+1, and θ i 1,j+1. This method is implicit: The value of any one of the three quantities at t j+1 cannot be calculated unless the other two are known. See exhibit (b) on p c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 648

37 Implicit Methods (continued) Equation (75) can be rearranged as θ i 1,j+1 (2 + γ) θ i,j+1 + θ i+1,j+1 = γθ i,j, where γ ( x) 2 /(D t). This equation is unconditionally stable. Suppose the boundary conditions are given at x = x 0 and x = x N+1. After θ i,j has been calculated for i = 1, 2,..., N, the values of θ i,j+1 at time t j+1 can be computed as the solution to the following tridiagonal linear system, c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 649

38 Implicit Methods (continued) a a a a a θ 1,j+1 θ 2,j+1 θ 3,j θ N,j+1 = γθ 1,j θ 0,j+1 γθ 2,j γθ 3,j γθ N 1,j γθ N,j θ N+1,j+1, where a 2 γ. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 650

39 Implicit Methods (concluded) Tridiagonal systems can be solved in O(N) time and O(N) space. The matrix above is nonsingular when γ 0. A square matrix is nonsingular if its inverse exists. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 651

40 Crank-Nicolson Method Take the average of explicit method (74) on p. 642 and implicit method (75) on p. 648: θ i,j+1 θ i,j = 1 2 ( t D θ i+1,j 2θ i,j + θ i 1,j ( x) 2 + D θ i+1,j+1 2θ i,j+1 + θ i 1,j+1 ( x) 2 ). After rearrangement, γθ i,j+1 θ i+1,j+1 2θ i,j+1 + θ i 1,j+1 2 = γθ i,j + θ i+1,j 2θ i,j + θ i 1,j 2. This is an unconditionally stable implicit method with excellent rates of convergence. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 652

41 Stencil x i+1 x i x i+1 t j t j+1 c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 653

42 Numerically Solving the Black-Scholes PDE See text. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 654

43 Monte Carlo Simulation a Monte Carlo simulation is a sampling scheme. In many important applications within finance and without, Monte Carlo is one of the few feasible tools. When the time evolution of a stochastic process is not easy to describe analytically, Monte Carlo may very well be the only strategy that succeeds consistently. a A top 10 algorithm according to Dongarra and Sullivan (2000). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 655

44 The Big Idea Assume X 1, X 2,..., X n have a joint distribution. θ E[ g(x 1, X 2,..., X n ) ] for some function g is desired. We generate ( ) x (i) 1, x(i) 2,..., x(i) n, 1 i N independently with the same joint distribution as (X 1, X 2,..., X n ). Set Y i g ( ) x (i) 1, x(i) 2,..., x(i) n. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 656

45 The Big Idea (concluded) Y 1, Y 2,..., Y N are independent and identically distributed random variables. Each Y i has the same distribution as Y g(x 1, X 2,..., X n ). Since the average of these N random variables, Y, satisfies E[ Y ] = θ, it can be used to estimate θ. The strong law of large numbers says that this procedure converges almost surely. The number of replications (or independent trials), N, is called the sample size. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 657

46 Accuracy The Monte Carlo estimate and true value may differ owing to two reasons: 1. Sampling variation. 2. The discreteness of the sample paths. a The first can be controlled by the number of replications. The second can be controlled by the number of observations along the sample path. a This may not be an issue if the derivative only requires discrete sampling along the time dimension. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 658

47 Accuracy and Number of Replications The statistical error of the sample mean Y of the random variable Y grows as 1/ N. Because Var[ Y ] = Var[ Y ]/N. In fact, this convergence rate is asymptotically optimal by the Berry-Esseen theorem. So the variance of the estimator Y can be reduced by a factor of 1/N by doing N times as much work. This is amazing because the same order of convergence holds independently of the dimension n. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 659

48 Accuracy and Number of Replications (concluded) In contrast, classic numerical integration schemes have an error bound of O(N c/n ) for some constant c > 0. n is the dimension. The required number of evaluations thus grows exponentially in n to achieve a given level of accuracy. The curse of dimensionality. The Monte Carlo method, for example, is more efficient than alternative procedures for securities depending on more than one asset, the multivariate derivatives. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 660

49 Monte Carlo Option Pricing For the pricing of European options on a dividend-paying stock, we may proceed as follows. Stock prices S 1, S 2, S 3,... at times t, 2 t, 3 t,... can be generated via S i+1 = S i e (µ σ2 /2) t+σ t ξ, ξ N(0, 1) (76) when ds/s = µ dt + σ dw. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 661

50 Monte Carlo Option Pricing (continued) If we discretize ds/s = µ dt + σ dw, we will obtain S i+1 = S i + (µ σ 2 /2) t + σ t ξ. But this is locally normally distributed, not lognormally, hence biased. a In practice, this is not expected to be a major problem as long as t is sufficiently small. a Contributed by Mr. Tai, Hui-Chin (R ) on April 22, c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 662

51 Monte Carlo Option Pricing (concluded) Non-dividend-paying stock prices in a risk-neutral economy can be generated by setting µ = r. 1: C := 0; 2: for i = 1, 2, 3,..., m do 3: P := S e (r σ2 /2) T +σ T ξ ; 4: C := C + max(p X, 0); 5: end for 6: return Ce rt /m; Pricing Asian options is easy (see text). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 663

52 How about American Options? Standard Monte Carlo simulation is inappropriate for American options because of early exercise (why?). It is difficult to determine the early-exercise point based on one single path. But Monte Carlo simulation can be modified to price American options with small biases (p. 709ff). a a Longstaff and Schwartz (2001). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 664

53 Delta and Common Random Numbers In estimating delta, it is natural to start with the finite-difference estimate rτ E[ P (S + ɛ) ] E[ P (S ɛ) ] e. 2ɛ P (x) is the terminal payoff of the derivative security when the underlying asset s initial price equals x. Use simulation to estimate E[ P (S + ɛ) ] first. Use another simulation to estimate E[ P (S ɛ) ]. Finally, apply the formula to approximate the delta. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 665

54 Delta and Common Random Numbers (concluded) This method is not recommended because of its high variance. A much better approach is to use common random numbers to lower the variance: [ ] P (S + ɛ) P (S ɛ) e rτ E. 2ɛ Here, the same random numbers are used for P (S + ɛ) and P (S ɛ). This holds for gamma and cross gammas (for multivariate derivatives). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 666

55 Gamma The finite-difference formula for gamma is [ P (S + ɛ) 2 P (S) + P (S ɛ) e rτ E ɛ 2 ]. For a correlation option with multiple underlying assets, the finite-difference formula for the cross gammas 2 P (S 1, S 2,... )/( S 1 S 2 ) is: [ P e rτ (S1 + ɛ 1, S 2 + ɛ 2 ) P (S 1 ɛ 1, S 2 + ɛ 2 ) E 4ɛ 1 ɛ 2 P (S 1 + ɛ 1, S 2 ɛ 2 ) + P (S 1 ɛ 1, S 2 ɛ 2 ) ]. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 667

56 Gamma (concluded) Choosing an ɛ of the right magnitude can be challenging. If ɛ is too large, inaccurate Greeks result. If ɛ is too small, unstable Greeks result. This phenomenon is sometimes called the curse of differentiation. Need formulas for Greeks which are integrals (thus avoiding ɛ, finite differences, and resimulation). a a Lyuu and Teng (R ) (2008). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 668

57 Biases in Pricing Continuously Monitored Options with Monte Carlo We are asked to price a continuously monitored up-and-out call with barrier H. The Monte Carlo method samples the stock price at n discrete time points t 1, t 2,..., t n. A sample path S(t 0 ), S(t 1 ),..., S(t n ) is produced. Here, t 0 = 0 is the current time, and t n = T is the expiration time of the option. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 669

58 Biases in Pricing Continuously Monitored Options with Monte Carlo (continued) If all of the sampled prices are below the barrier, this sample path pays max(s(t n ) X, 0). Repeating these steps and averaging the payoffs yield a Monte Carlo estimate. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 670

59 1: C := 0; 2: for i = 1, 2, 3,..., m do 3: P := S; hit := 0; 4: for j = 1, 2, 3,..., n do 5: P := P e (r σ2 /2) (T /n)+σ (T /n) ξ ; 6: if P H then 7: hit := 1; 8: break; 9: end if 10: end for 11: if hit = 0 then 12: C := C + max(p X, 0); 13: end if 14: end for 15: return Ce rt /m; c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 671

60 Biases in Pricing Continuously Monitored Options with Monte Carlo (continued) This estimate is biased. Suppose none of the sampled prices on a sample path equals or exceeds the barrier H. It remains possible for the continuous sample path that passes through them to hit the barrier between sampled time points (see plot on next page). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 672

61 H c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 673

62 Biases in Pricing Continuously Monitored Options with Monte Carlo (concluded) The bias can certainly be lowered by increasing the number of observations along the sample path. However, even daily sampling may not suffice. The computational cost also rises as a result. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 674

63 Brownian Bridge Approach to Pricing Barrier Options We desire an unbiased estimate efficiently. So the above-mentioned payoff should be multiplied by the probability p that a continuous sample path does not hit the barrier conditional on the sampled prices. This methodology is called the Brownian bridge approach. Formally, we have p Prob[ S(t) < H, 0 t T S(t 0 ), S(t 1 ),..., S(t n ) ]. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 675

64 Brownian Bridge Approach to Pricing Barrier Options (continued) As a barrier is hit over a time interval if and only if the maximum stock price over that period is at least H, [ ] p = Prob max S(t) < H S(t 0), S(t 1 ),..., S(t n ). 0 t T Luckily, the conditional distribution of the maximum over a time interval given the beginning and ending stock prices is known. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 676

65 Brownian Bridge Approach to Pricing Barrier Options (continued) Lemma 19 Assume S follows ds/s = µ dt + σ dw and define [ ] 2 ln(x/s(t)) ln(x/s(t + t)) ζ(x) exp. σ 2 t (1) If H > max(s(t), S(t + t)), then [ ] Prob max S(u) < H t u t+ t S(t), S(t + t) = 1 ζ(h). (2) If h < min(s(t), S(t + t)), then [ ] Prob min S(u) > h t u t+ t S(t), S(t + t) = 1 ζ(h). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 677

66 Brownian Bridge Approach to Pricing Barrier Options (continued) Lemma 19 gives the probability that the barrier is not hit in a time interval, given the starting and ending stock prices. For our up-and-out call, choose n = 1. As a result, p = [ 1 exp 2 ln(h/s(0)) ln(h/s(t )) σ 2 T ], if H > max(s(0), S(T )), 0, otherwise. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 678

67 Brownian Bridge Approach to Pricing Barrier Options (continued) 1: C := 0; 2: for i = 1, 2, 3,..., m do 3: P := S e (r q σ2 /2) T +σ T ξ( ) ; 4: if (S < H and P < H) or { (S > H and [ P > H) then 5: C := C+max(P X, 0) 1 exp 6: end if 7: end for 8: return Ce rt /m; 2 ln(h/s) ln(h/p ) σ 2 T ]} ; c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 679

68 Brownian Bridge Approach to Pricing Barrier Options (concluded) The idea can be generalized. For example, we can handle more complex barrier options. Consider an up-and-out call with barrier H i time interval (t i, t i+1 ], 0 i < n. for the This option thus contains n barriers. It is a simple matter of multiplying the probabilities for the n time intervals properly to obtain the desired probability adjustment term. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 680

69 Variance Reduction The statistical efficiency of Monte Carlo simulation can be measured by the variance of its output. If this variance can be lowered without changing the expected value, fewer replications are needed. Methods that improve efficiency in this manner are called variance-reduction techniques. Such techniques become practical when the added costs are outweighed by the reduction in sampling. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 681

70 Variance Reduction: Antithetic Variates We are interested in estimating E[ g(x 1, X 2,..., X n ) ], where X 1, X 2,..., X n are independent. Let Y 1 and Y 2 be random variables with the same distribution as g(x 1, X 2,..., X n ). Then [ Y1 + Y 2 Var 2 ] = Var[ Y 1 ] 2 + Cov[ Y 1, Y 2 ]. 2 Var[ Y 1 ]/2 is the variance of the Monte Carlo method with two (independent) replications. The variance Var[ (Y 1 + Y 2 )/2 ] is smaller than Var[ Y 1 ]/2 when Y 1 and Y 2 are negatively correlated. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 682

71 Variance Reduction: Antithetic Variates (continued) For each simulated sample path X, a second one is obtained by reusing the random numbers on which the first path is based. This yields a second sample path Y. Two estimates are then obtained: One based on X and the other on Y. If N independent sample paths are generated, the antithetic-variates estimator averages over 2N estimates. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 683

72 Variance Reduction: Antithetic Variates (continued) Consider process dx = a t dt + b t dt ξ. Let g be a function of n samples X 1, X 2,..., X n the sample path. on We are interested in E[ g(x 1, X 2,..., X n ) ]. Suppose one simulation run has realizations ξ 1, ξ 2,..., ξ n for the normally distributed fluctuation term ξ. This generates samples x 1, x 2,..., x n. The estimate is then g(x), where x (x 1, x 2..., x n ). c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 684

73 Variance Reduction: Antithetic Variates (concluded) The antithetic-variates method does not sample n more numbers from ξ for the second estimate g(x ). Instead, generate the sample path x (x 1, x 2..., x n) from ξ 1, ξ 2,..., ξ n. Compute g(x ). Output (g(x) + g(x ))/2. Repeat the above steps for as many times as required by accuracy. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 685

74 Variance Reduction: Conditioning We are interested in estimating E[ X ]. Suppose here is a random variable Z such that E[ X Z = z ] can be efficiently and precisely computed. E[ X ] = E[ E[ X Z ] ] by the law of iterated conditional expectations. Hence the random variable E[ X Z ] is also an unbiased estimator of E[ X ]. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 686

75 Variance Reduction: Conditioning (concluded) As Var[ E[ X Z ] ] Var[ X ], E[ X Z ] has a smaller variance than observing X directly. First obtain a random observation z on Z. Then calculate E[ X Z = z ] as our estimate. There is no need to resort to simulation in computing E[ X Z = z ]. The procedure can be repeated a few times to reduce the variance. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 687

76 Control Variates Use the analytic solution of a similar yet simpler problem to improve the solution. Suppose we want to estimate E[ X ] and there exists a random variable Y with a known mean µ E[ Y ]. Then W X + β(y µ) can serve as a controlled estimator of E[ X ] for any constant β. However β is chosen, W remains an unbiased estimator of E[ X ] as E[ W ] = E[ X ] + βe[ Y µ ] = E[ X ]. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 688

77 Control Variates (continued) Note that Var[ W ] = Var[ X ] + β 2 Var[ Y ] + 2β Cov[ X, Y ], (77) Hence W is less variable than X if and only if β 2 Var[ Y ] + 2β Cov[ X, Y ] < 0. (78) c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 689

78 Control Variates (concluded) The success of the scheme clearly depends on both β and the choice of Y. For example, arithmetic average-rate options can be priced by choosing Y to be the otherwise identical geometric average-rate option s price and β = 1. This approach is much more effective than the antithetic-variates method. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 690

79 Choice of Y In general, the choice of Y is ad hoc, and experiments must be performed to confirm the wisdom of the choice. Try to match calls with calls and puts with puts. a On many occasions, Y is a discretized version of the derivative that gives µ. Discretely monitored geometric average-rate option vs. the continuously monitored geometric average-rate option given by formulas (30) on p For some choices, the discrepancy can be significant, such as the lookback option. b a Contributed by Ms. Teng, Huei-Wen (R ) on May 25, b Contributed by Mr. Tsai, Hwai (R ) on May 12, c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 691

80 Optimal Choice of β Equation (77) on p. 689 is minimized when β = Cov[ X, Y ]/Var[ Y ], which was called beta in the book. For this specific β, Var[ W ] = Var[ X ] Cov[ X, Y ]2 Var[ Y ] = ( 1 ρ 2 X,Y ) Var[ X ], where ρ X,Y is the correlation between X and Y. The stronger X and Y are correlated, the greater the reduction in variance. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 692

81 Optimal Choice of β (continued) For example, if this correlation is nearly perfect (±1), we could control X almost exactly. Typically, neither Var[ Y ] nor Cov[ X, Y ] is known. Therefore, we cannot obtain the maximum reduction in variance. We can guess these values and hope that the resulting W does indeed have a smaller variance than X. A second possibility is to use the simulated data to estimate these quantities. How to do it efficiently in terms of time and space? c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 693

82 Optimal Choice of β (concluded) Observe that β has the same sign as the correlation between X and Y. Hence, if X and Y are positively correlated, β < 0, then X is adjusted downward whenever Y > µ and upward otherwise. The opposite is true when X and Y are negatively correlated, in which case β > 0. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 694

83 A Pitfall A potential pitfall is to sample X and Y independently. In this case, Cov[ X, Y ] = 0. Equation (77) on p. 689 becomes Var[ W ] = Var[ X ] + β 2 Var[ Y ]. So whatever Y is, the variance is increased! c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 695

84 Problems with the Monte Carlo Method The error bound is only probabilistic. The probabilistic error bound of N does not benefit from regularity of the integrand function. The requirement that the points be independent random samples are wasteful because of clustering. In reality, pseudorandom numbers generated by completely deterministic means are used. Monte Carlo simulation exhibits a great sensitivity on the seed of the pseudorandom-number generator. c 2010 Prof. Yuh-Dauh Lyuu, National Taiwan University Page 696