We present some results of convergence for a minimum contrast estimator in
a problem of change-points estimation. Here, we consider that the changes a
ffect the marginal distribution of a sequence of random variables. We only
consider parametric models, but the results are obtained under very general
conditions. We show that the estimated configuration of changes converges
to the true configuration, and we show that the rate of convergence does no
t depend on the dependance structure of the process: we obtain the same rat
e for strongly mixing and strongly dependent processes. When the number of
changes is unknown, it is estimated by minimizing a penalized contrast func
tion. Some examples of application to real data are given. (C) 1999 Publish
ed by Elsevier Science B.V. All rights reserved.