Bayesian inference and forecasting of time series under the different loss functions | Posted on:2006-03-09 | Degree:Ph.D | Type:Dissertation | University:The University of North Carolina at Charlotte | Candidate:Chen, Jian | Full Text:PDF | GTID:1450390005498894 | Subject:Statistics | Abstract/Summary: | | We consider the different loss functions that are appropriate for the Bayesian analysis of some time series models. The Bayes inference and forecasting under these loss functions are given.;For the autoregressive model, with the Normal-Gamma and Jeffreys' priors, the posteriors are found and Bayes estimates for the parameters in the model under the different loss functions are derived, the probability density function of k-step ahead Bayes prediction is derived in a concise matrix format. In particular, Bayes estimates of the one-step ahead forecasting under the different loss functions are given. We provide the practical k-step ahead Bayesian forecasting under these loss functions. The Bayes estimates and one-step ahead and two-step ahead forecasting results under these loss functions are calculated. Under the Normal-Gamma and Jeffreys' priors, Wolfer sunspot numbers data is used to illustrate Bayes inferences and forecasts to the real life data.;For the moving average model, under the Gamma-Normal and Jeffreys' priors, based on the approximate likelihood function, the posteriors and one-step ahead forecasting probability density function are derived. Then we obtain the Bayes estimates for the parameters and predictive inferences for moving average processes under the different loss functions. | Keywords/Search Tags: | Loss functions, Time series, Bayes estimates for the parameters, Bayesian, Forecasting, Moving average | | Related items |
| |
|