STA2570 Winter 2023 Assignment 1
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
STA2570 Winter 2023
Assignment 1
Feb 11, 2023
1. Let C be the Frank copula with parameter θ = 5. You may look up the definition here:
https://en.wikipedia.org/wiki/Copula_ (probability_theory)
It can be readily implemented using available software packages which you can use. Let (X, Y) be a bivariate random vector such that (i) its copula is C and (ii) both the marginal distributions of X and Y are both N (0, 1).
(a) Using a direct Monte Carlo simulation, determine the correlation
ρ = Cov(X, Y )
^Var(X) ^Var(Y ) .
between X and Y . Using a preliminary study if necessary, determine a sample size needed such that the 99% confidence interval is shorter than 0 .01, and report one such confidence interval.
(b) Let F一1 be the common quantile function of X and Y . For q 兰 (0, 1), let λ(q) = P(Y > F一1 (q)EX > F一1 (q)).
Estimate λ(q), as accurately as you can, for q 兰 [0.5, 0.6, 0.7, 0.8, 0.9, 0.95}. This is related to the concept of (upper) tail dependence.
(c) Let ρ be the correlation in (a) [you may use the estimate you get], and let ( , ) be a bivariate normal random vector with zero mean and covariance matrix
Σ = ┌ρ(1) 1(ρ)┐ .
Find or estimate the corresponding values
(q) = P( > F一21 (q)E > F一11 (q)),
for the values of q in (b). Comment on your results.
2. Consider a geometric Brownian motion
dSt = µStdt + σStdWt, S0 = s0 ,
where µ = 0.3, σ = 0.4 and s0 = 10.
Simulate 100000 discretized Brownian paths over [0, 1] with time step δt = 2一9 . Note that we have the “exact solution” St = S0 exp ╱(µ 2 σ 2 )t + σWt、.
For 5 different step sizes: ∆t = 2p一1 δt, 1 k p k 5, apply both the Euler-Maruyama scheme (call the output EM ) and the Milstein scheme (call the output M ).
Plot the following graphs in log-log scale:
❼ Sample average of ES1 2 1(EM)E versus ∆t.
❼ Sample average of ES1 2 1(M)E versus ∆t.
❼ EPS1 2 sample average of 1(EM)E versus ∆t.
❼ EPS1 2 sample average of 1(M)E versus ∆t.
Fit a straight line for each graph and discuss your results in relation to the weak and strong convergence criteria.
Note: You may refer to the following review paper for a similar example. See in particular Figure 4.
❼ Higham, D. J. (2001). An algorithmic introduction to numerical simulation of stochas-
tic differential equations. SIAM Review, 43(3), 525–546.
3. Consider, under the Black-Scholes model
dSt = rStdt + σStdWt , 0 k t k T (under Q),
a digital exotic option with payoff
CT = 1[ min Stk < H},
where t1 < | | | < tm is a discrete collection of times before maturity. In this problem we want to estimate the price C0 at time 0 using importance sampling. In particular, we consider an experiment to find an “optimal change of measure” .
(a) If the payoff is 1[min0≤t≤T St < H} (continuous monitoring instead of discrete moni-
toring), find the theoretical value of the option. Hint: Look up the distribution of the minimum of a Brownian motion over a time interval, and use the risk-neutral pricing formula.
(b) We use the following parameters (same as those used on p.266 of Glasserman):
- T = 0.25, r = 0.05, σ = 0.15, S0 = 95, H = 85, tk = T , m = 50.
For θ > 0, let Qθ be the measure such that Wt becomes a Brownian motion with negative drift:
Wt = 2θt + Wtθ ,
where Wθ is a Qθ -Brownian motion. Derive the likelihood ratio (from Girsanov’s theorem) and the corresponding importance sampling estimator for the price C0 at t = 0.
(c) Sample 100000 standard Brownian paths. Note that we may reuse the same paths for different values of θ simply by adding a drift. Estimate the standard error of the importance sampling estimator over a reasonable range of θ . Plot the (estimated) variance as a function of θ . What is the optimal θ?
2023-02-19