1: \begin{abstract}

2: The multi-level Monte Carlo method proposed by M.~Giles~(2008)

3: approximates the expectation of some functionals applied to a

4: stochastic process with optimal order of convergence for the

5: mean-square error. In this paper, a modified multi-level Monte Carlo

6: estimator is proposed with significantly reduced computational

7: costs. As the main result, it is proved that the modified estimator

8: reduces the computational costs asymptotically by a factor

9: $(p/\alpha)^2$ if weak approximation methods of orders $\alpha$ and

10: $p$ are applied in case of computational costs growing with same

11: order as variances decay.

12: \end{abstract}

13: