Methods for the detection of intervention effects and change-points at known or unknown time points in sequential data are well established under classical Gaussian assumptions. However, such techniques are far from being optimal or even not applicable when the data are discrete, as in the situation of count time series arising for instance in epidemiology, heavy-tailed, as for instance in finance, or very skewed, as for instance in hydrology. The talk proposes some alternatives to traditional t-tests and its offsprings like the cumulative sums (CUSUM) test which we consider more suitable for non-Gaussian data scenarios as those mentioned before.

Einladung von Prof. Didelez