Title: Parametric predictive bootstrap
Authors: Abdulrahman Aldawsari - Durham University (United Kingdom) [presenting]
Abstract: Bootstrap methods are used to quantify the uncertainty of sample estimates, they have been applied to a wide range of statistical problems due to their simplicity and efficiency in giving good estimates. There are two main bootstrap methods: parametric and nonparametric. The parametric bootstrap method uses available data to estimate the parameters of the assumed distribution and then generates a number of parametric bootstrap samples from the assumed distribution with the estimated parameters. A new bootstrap method is proposed, especially for predictive inference, the parametric predictive bootstrap, which does use an assumed parametric model. It is evaluated in a variety of scenarios that have been used with other bootstrap methods in order to investigate its performance in estimation and prediction inference. The proposed bootstrap method is compared to different types of bootstrap methods in terms of the coverage probability through simulations. Confidence intervals and prediction intervals based on the bootstrap technique are used to examine the PP-B's performance in estimation and prediction inference. The explicitly predictive nature of PP-B provides good performance for predictive inference.