On a multidimensional general bootstrap for empirical estimator of continuous-time semi-Markov kernels with applications |
| |
Authors: | Salim Bouzebda Chrysanthi Papamichail Nikolaos Limnios |
| |
Affiliation: | Laboratoire de Mathématiques Appliquées de Compiègne, Sorbonne Universités, Université de Technologie de Compiègne, France |
| |
Abstract: | The present paper introduces a general notion and presents results of bootstrapped empirical estimators of the semi-Markov kernels and of the conditional transition distributions for semi-Markov processes with countable state space, constructed by exchangeably weighting the sample. Our proposal provides a unification of bootstrap methods in the semi-Markov setting including, in particular, Efron's bootstrap. Asymptotic properties of these generalised bootstrapped empirical distributions are obtained, under mild conditions by a martingale approach. We also obtain some new results on the weak convergence of the empirical semi-Markov processes. We apply these general results in several statistical problems such as the construction of confidence bands and the goodness-of-fit tests where the limiting distributions are derived under the null hypothesis. Finally, we introduce the quantile estimators and their bootstrapped versions in the semi-Markov framework and we establish their limiting laws by using the functional delta methods. Our theoretical results and numerical examples by simulations demonstrate the merits of the proposed techniques. |
| |
Keywords: | Semi-Markov process semi-Markov kernel empirical estimator functional central limit theorem semimartingale bootstrap exchangeable bootstrap confidence interval goodness-of-fit test |
|
|