Hill estimator and extreme quantile estimator for functionals of approximated stochastic processes

We study the effect of approximation errors in assessing the extreme behaviour of univariate functionals of random objects. We build our framework into a general setting where estimation of the extreme value index and extreme quantiles of the functional is based on some approximated value instead of the true one. As an example, we consider the effect of discretisation errors in computation of the norms of paths of stochastic processes. In particular, we quantify connections between the sample size $n$ (the number of observed paths), the number of the discretisation points $m$, and the modulus of continuity function $\phi$ describing the path continuity of the underlying stochastic process. As an interesting example fitting into our framework, we consider processes of form $Y(t) = \mathcal{R}Z(t)$, where $\mathcal{R}$ is a heavy-tailed random variable and the increments of the process $Z$ have lighter tails compared to $\mathcal{R}$.