Direct and filtered bootstrap in time series models.

摘要

This work aims at the implementation of 2 kinds of bootstrap in the estimation of AR92) autoregressive parameters and comparision between them. The implementations are called "direct bootstrap" and "filtered bootstrap". The direct bootstrap consists of a multivariate reseample of the series itself and with delays, seeking to replicate the autoregression matrices. The filtered bootstrap, tested here by way of 2 different implementations, uses a reseample of the estimated residuals of the adjusted series to replicate the series itself. Hypothesis tests are proposed for the nullity of the parameters.