Bayesian analysis of structural change in trend

Zheng, P, 2002. Bayesian analysis of structural change in trend. PhD, Nottingham Trent University.

Preview

Text 10290345.pdf - Published version Download (21MB) | Preview

Abstract

In recent years there has been a lot interest in studying the structural changes in time series, especially in economic time series. Statistical methods have been developed to deal with this kind of problem. Bayesian methods are becoming more and more popular in this field.

This thesis focuses on two kinds of structural change, abrupt structural change which happens at some point and the gradual structural change which happens over a period of time. For the former structural change, we discuss the two-phase model and the structural break model; for the latter structural change, we discuss the smooth transition model. We address the problem of parameter estimation for these models using a Bayesian approach. We derive expressions for the posterior densities for parameters which are used to make inference of the parameters and posterior model probabilities which are used to compare models. We also discuss the double smooth transition model which has two smooth transition components. Markov chain Monte Carlo methods are used to estimate these models, including parameter estimation and model selection. Models are fitted with their predictive means. We illustrate our approaches with empirical examples such as the British industrial production index, the US economic time series from Nelson and Plosser (1982) and the global average temperature series.

Finally we apply the reversible jump Markov chain Monte Carlo method to a structural break model which has unknown number of structural break points. The posterior model probabilities are obtained for models with different structural break points and posterior densities for parameters in the preferred model (with the biggest estimated posterior model probability) are also obtained. We fit the model to two US economic times series, the US real GNP series and the US consumer price index, with the number of structural break points selected by our algorithm automatically.

Item Type:	Thesis
Creators:	Zheng, P.
Date:	2002
ISBN:	9781369325942
Identifiers:	Number Type PQ10290345 Other
Divisions:	Schools > School of Social Sciences
Record created by:	Laura Ward
Date Added:	12 Jul 2021 15:06
Last Modified:	24 Jul 2024 14:11
URI:	https://irep.ntu.ac.uk/id/eprint/43417

Actions (login required)

Edit View

Statistics

Views

Downloads