Table of Contents:
Topic of the doctoral thesis is the theoretical
investigation of methods of change-point analysis for application
to telecommunication data.
The models used here are motivated by models which are used in
telecommunications. In particular for market share investigations
related to phoned minutes linear models are appropriate. The error
terms are frequently not independent as often assumed in
theoretical investigations but correlated. We generalize
procedures of a posteriori change-point analysis for linear models
with independent errors to such allowing correlated error terms.
Another important very general class of models is the model class
of State Space models. State Space models are used in particular
for prognoses of traffic volume in telecommunications. New
procedures for a posteriori change-point analysis are developed
for this model class.
Concerning sequential procedures of change-point analysis we
present well-known practice-oriented procedures and transfer these
on linear models as well as on State Space models.
Results obtained theoretically are subject of simulation studies.
All programs, which are used for these, are written in the
statistical programming language R.
The methods examined here are applied to original data of Deutsche
Telekom in an application study. This application study is not
part of the thesis for data security reasons. On the one hand the
conception of early warning systems is discussed, which examine
observations sequentially for structural breaks (deviations from a
given model). On the other hand we regard analysis systems, which
examine a set of observations in retrospect (''a posteriori'') for
existing change-points. The procedures supply valuable hints about
structural breaks in observed data. Some of the procedures are
used by Deutsche Telekom for automated monitoring of market shares
as well as other characteristics.