On asymptotic errors in discretization of processes (Q1394519)

The authors study the rate at which the difference \(X_t^n=X_t-X_{[ng]/n}\) between a càdlàg real-valued stochastic process \(X\) and its time-discretization converges to zero. It is known that for continuous semimartingales under natural conditions this rate equals \(n^{-1/2}\). To treat the discrete case it is necessary (for reasons given in the paper) to consider the integrated error processes \(Y_t^n=\int^t_0 X^n_s\,ds\) and \(Z_t^{n,p}= \int^t_0 |X_s^n|^p\,ds\) for \(0<p<\infty\). Among others, under appropriate assumptions the tightness of the sequences \(n\) \(\sup_{s\leq t}|Y^n_s|\) and \(n\sup_{s\leq t} Z_s^{n,p}\) and the weak convergence of the discretized processes \(nY^n_{[nt]/n}\) and \(nZ^{n,p}_{[nt]/n}\) are proved. As a by-product, a generalization of Itô's formula for functions that are not twice continuously differentiable and which is of interest for its own is given.