Bootstrap and statistical inference for phase-type distributions

Specialeforsvar ved Malte Lindholm Stolte Rasmussen

Titel: Bootstrap and statistical inference for phase-type distributions 

 

Abstract: This thesis deals with statistical inference on phase-type distributions. As one of the main contributions, an approach for obtaining well-defined sample distributions for the estimated parameter values is provided and implemented. This approach involves a combination of nonparametric bootstrap with the EM algorithm and provides an alternative for deriving (empirical) confidence intervals for phase-type distributions, which may be used in statistical inference. In addition, this work focuses on the denseness of continuous phase-type distributions in the positive real axis. A numerical study of the denseness property is carried out, and a new way of approximating non-negative continuous distributions with phase-type distributions is provided. Keywords: phase-type distributions; EM algorithm; Bootstrap

 

 

Vejleder: Mogens Bladt
Censor:   Bo Friis Nielsen, DTU