|
Organizers |
Short-term prediction of time series using semiparametric Bayesian techniques
by
Kaushik Ghosh
Department of Statistics, George Washington University, Washington, D.C.
We present Bayesian local-linear and local-quadratic models for short-term prediction of time series. Dirichlet process priors are assumed on the distribution of the slope and acceleration terms r espectively to make the model flexible and to accommodate various shapes of the series. Markov chain Monte Carlo techniques are used to obtain predicted values along with prediction intervals. We illustrate the proposed models using mortality data on some common occurring cancers in the U.S. and discuss how one would choose prior parameters to balance model smoothness with flexibility.
Date received: September 13, 2005
Copyright © 2005 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # caqt-67.