Parameters Estimate And Bootstrap Confidence Intervals Of Generalized Exponential Family ARMA Models

The objective of this paper is to estimate the parameters of generalized exponential family ARMA models and construct the confidence intervals of the parameters. In this paper we divided the problem into continuous type situation and discrete type situation. Under both situations, we use Fisher Scoring algorithm to estimate the parameters and gain the formula of direction vectors in the algorithm. Under the continuous situation we use Wild Bootstrap to construct the confidence intervals of the parameters, then compare the simulation results to the Remain Bootstrap, then get the conclusion that the Wild Bootstrap is more quickly and more accurate; Under the discrete situation the method of Moving Blocks Bootstrap gives satisfied results for both simulative and true data.