A new parameterization for the drift-free simulation in the Libor market model (Q2341004)

In this paper, the authors present an alternative parametrized algorithm to simulate the dynamics of Libor market models. The procedure is based on a system of stochastic differential equations proposed in [\textit{P. Glasserman} and \textit{X. Zhao}, Finance Stoch. 4, No. 1, 35--68 (2000; Zbl 0947.60050)]. More concretely, consider a fixed tenure structure \(\mathcal{T} = \{ T_0,T_1,\ldots,T_{N+1} \}\) with \(T_0 = 0\), \(T_j < T_k\), \(0 \leq j < k \leq N+1\), and the corresponding accruals \(\delta_j = T_{j+1} - T_j\), \(0 \leq j \leq N\). The authors show that the forward Libor rates are given by \[ L_j(t) = \frac{X_j(t)}{\delta_j \sum_{l=j+1}^{N+1} X_l(t)}, \] where the \(X_j\) are martingales, which are given by the solutions \(u_j\) of certain stochastic differential equations, and certain martingales \(M_j\). For their algorithm, the authors propose Euler discretizations \(\hat{u}_j\) for the solutions \(u_j\) of the stochastic differential equations, which provides approximations \(\hat{X}_j\), \(\hat{M}_j\) and \(\hat{L}_j\) of \(X_j\), \(M_j\) and \(L_j\), and they prove that \(0 < \hat{M}_{j+1} < \hat{M}_j\) and \(\hat{L}_j > 0\). Furthermore, as the martingale property for \(X_j\) can be lost with the discretization \(\hat{X}_j\), the authors provide an approximation for a martingale adjustment suggested in [Glasserman and Zhao, loc. cit.]. Model calibration and numerical results are considered as well.