STA 356

STA 356 Time Series/Forecasting

Link to:

Mike West's online "tutorial"
announcements
Splus source
Notes on AR models
Some of the EEG data
Some Strontium data
Some S-Plus code for reference AR analysis
Some S-Plus code to simulate AR(2) processes
New EEG data for Homework 12.
homework problems
Assignment 12 Final assessments

STA 356 TIME SERIES/FORECASTING. Credits: 0.00, Hours: 3.0. 
Section 01. Call No. 138566. Limit: 18. Instructor: Lavine and West. 
MWF, 11:50AM- 12:40 in Old Chemistry, room 025.

We will cover some basic exploratory data analysis (autocorrelations, periodogram etc.), autoregressive/moving average models, and inference in dynamic state space models. Dynamic state space models are time series models based on assuming that there are some underlying parameters (like mean level mu[t] of y[t]; rate of change d[t]; maybe seasonal components etc) which describe underlying structure and which "evolve" between time periods. We will discuss Bayesian inference in such models. Discussion of dynamic state space models will include basic concepts and posterior inference for standard normal dynamic linear models (NDLM), posterior simulation based on forward filtering backward smoothing, extensions to conditionally NDLM models and general non-normal, non-linear models.

Students are encouraged to bring applications arising from their own research to class.

Class Format:

Lectures on Monday and Wednesday, student presentation of assigned problems on Fridays.

Grading:

Based on weekly homework (typically one or two problems) and an in-class presentation of a worked out problem.

Textbooks:

West, M. and Harrison, P.J. (1997), Bayesian Forecasting and Dynamic Models, Springer-Verlag, (2nd Edition). The book covers Bayesian inference in dynamic state space models.

Diggle, P., Time Series: A Biostatistical Introduction, Oxford University Press, ISBN 0-19-852226-6 (pbk). This book will serve as reference for basic concepts of time series analysis, i.e., various exploratory data analysis, basic ARMA modelling etc. Any alternative introductory time series text is fine.

Prerequisites

STA 214, 215, 242 and/or 244, or equivalent. Some prior computing experience is useful. We will assume that people are familiar with basic principles of Bayesian analysis, i.e., prior/likelihood/posterior, Bayes' theorem, basic conjugate models (normal/normal; beta/binomial; discrete).

Here's a simple "self test" to find out for yourself whether you have adequate background: Get a copy of the STA 215 text book Gelman, Carlin, Stern and Rubin, Bayesian Data Analysis. Chapman and Hall Look at the following problems: Chapter 2: 1,2,3,5,7,9,10 Just see if you would know how to approach these problems. If you do, then your background is adequate, if not, the class might be a little challenging for you. Don't worry if you have to look up things to solve them. That's fine. Only, you should know enough to know where to look up things etc.

Reading List

Some references in time series of at least peripheral interest

Have a look at Mike West's online "tutorial" on Bayesian time series analysis and forecasting.

michael@stat.duke.edu

last updated 1/29/97