1: \begin{abstract}

2: %In this paper we show that the Milstein scheme can be used to improve the

3: %convergence of the multilevel Monte Carlo method for scalar stochastic

4: %differential equations. Numerical results for Asian, lookback, barrier and

5: %digital options demonstrate that the computational cost to achieve a

6: %root-mean-square error of $\eps$ is reduced to $O(\eps^{-2})$. This is

7: %achieved through a careful construction of the multilevel estimator which

8: %computes the difference in expected payoff when using different numbers of

9: %timesteps.

10: 

11: %In this work we applies multilevel approach to Monte Carlo simulation of option pricing under a jump-diffusion SDE. In order to reduce

12: %the variance of the correction terms series, we adopt a jump-adapted Milstein

13: %discretisation scheme. For path-dependent cases, the estimators for payoffs is constructed using

14: %Brownian interpolation within the time grid. Numerical results for

15: %Asian, lookback, barrier and digital options show that the computational

16: %cost to achieve a root-mean-square error of $\eps$ is reduced to $O(\eps%

17: %^{-2})$, which is similar to the previous case of pure diffusion processes.

18: 

19: We investigate the extension of the multilevel Monte Carlo path simulation

20: method to jump-diffusion SDEs. We consider models

21: with finite rate activity , using a jump-adapted

22: discretisation in which the jump times are computed and added to the

23: standard uniform discretisation times. The key component in multilevel

24: analysis is the calculation of an expected payoff difference between a

25: coarse path simulation and a fine path simulation with twice as many

26: timesteps. If the Poisson jump rate is constant, the jump times are the same

27: on both paths and the multilevel extension is relatively straightforward,

28: but the implementation is more complex in the case of state-dependent jump

29: rates for which the jump times naturally differ.

30: \end{abstract}