- Thesis:
- Bachelor in Mathematics 50%
- Author:
- Miriam Maurer
- Title:
- Adaptive Laguerre density estimation for mixed Poisson models
- Supervisors:
- Sergio Brenner Miguel and Jan JOHANNES
- Abstract:
- In non-life insurances a commonly used model is given by the Poisson process model. Due to dependencies on different quantities the assumption that the intensity is a fixed parameter is not reasonable in many applications. Therefore it makes sense to interpret the intensity as a random variable on the positive real line, which leads us to the more general mixed Poisson process model. In this thesis the unknown density f of the random intensity is estimated using an iid sample of the value of a mixed Poisson process at a fixed observation time. This is done by a penalized projection approach using the Laguerre basis. The focus is to introduce a data-driven selection rule for an adequate dimension parameter in order to deal with the upcoming bias-variance conflict. An upper bound of the L2-risk is obtained and a lower bound is provided, which proves the optimality of the estimator. Further the procedure is illustrated via a monte-carlo simulation.
- References:
- F. Comte and V. Genon-Catalot. Adaptive laguerre density estimation for mixed Poisson models. Electronic Journal of Statistics, 9(1):1113–1149, 2015.
| Statistics Group | Institute for Mathematics | Faculty of Mathematics and Computer Science | University Heidelberg |