Abstract
In the thesis we proposed several different procedures, how to detect changes in sequentially coming data following a simple locations or linear regression model with i.i.d. errors. We assume that training data with no change are available and focus on changes in the llocation parameter and the regression coefficients, respectively.
We propose CUSUM type test procedures based on L1-residuals or on (studentized) M-residuals, such that the model is estimated only from the training data. We derived the limit distributions of these test procedures under the null hypothesis and show, that they are consistent against the alternatives.
The theoretical results are confirmed in simulations.