Bayesian Statistics: Dynamic Linear Modelling, 2003

Jonathan Rougier

The first lecture is on Monday 17 February. The remaining lectures follow twice-weekly from Wednesday 26 February. They occur on Monday 11.00 in CM107 and Wednesday 9.00 in CM105. Please come and get me from CM312 if I have forgotten to turn up for the lecture, or call me on extension 2361.

If you would like to talk to me about the course material please see me to arrange a time at the beginning or end of a lecture.

On this page you can find the following information.

Handouts

There is no one set book for these lectures. However, the material is covered in West, M. and J. Harrison, Bayesian Forecasting and Dynamic Models, Springer (2ed, 1997, 3ed now available). This book also provides the notation. There will be handouts distributed at the beginning of each lecture. These handouts will cover the core material. If you find mistakes in these handouts please let me know either in the lecture (hopefully!) or by e-mail to J.C.Rougier@durham.ac.uk.

The lectures and handouts are as follows.
17.02.03 Introduction: Example of a DLM in action postscript PDF
Core theory
26.02.03 Basic features, and forecasting postscript PDF
03.03.03 Assimilating new data postscript PDF
05.03.03 Filtering, or `backcasting' postscript PDF
Applied methods
10.03.03 Simple time-series models postscript PDF
12.03.03 Non-linear models postscript PDF
17.03.03 The discounting approach postscript PDF
19.03.03 Example using property data (incl. seasonal effetcs) postscript

Question sheets

Questions are appended to each handout, in the form of exercises. "Answers" will be available from here in due course... Here they are: postscript, or PDF.

Note that you can get the latest Bayes Linear solutions from David Wooff's Bayes linear statistics page.

Software

I have written some software to accompany these lectures, just in case you get the urge. It is written as a library of the R statistical software environment. Most of you should remember R from computer practicals in your previous years. The package "tsdlm" is installed on the University network, just type "library(tsdlm)" at the R prompt. If you want to use this software on your own computer, you will need the three packages tsdlm, uuobj and tensor. You can install these from the menu in the Windows version of R.

Once you have done "library(tsdlm)" to load the DLM package (it also loads the utility package "uuobj") you can get help using "?tsdlm". This help file contains all of the code to do the example in the first lecture. You can re-create the example of the final lecture by following the help page example on "?property".

Last modified: Thu May 1 11:08:22 BST 2003