Use this URL to cite or link to this record in EThOS:
Title: Bayesian inference on non-stationary data
Author: Amisano, Giovanni
ISNI:       0000 0001 3419 6812
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 1995
Availability of Full Text:
Access from EThOS:
Access from Institution:
This thesis argues in favour of Bayesian techniques for the analysis of non- stationary linear time series. The main motivations are to avoid using asymptotic results and to explicitly incorporate prior beliefs, where they exist. The properties of univariate and multivariate unit root models, and the available frequentist inferential results are described. Some problems in their applications are highlighted: the discrepancies between asymptotic and finite sample properties and the role of the deterministic components in determining the reference asymptotic distributions. The advantages and disadvantages of Bayesian techniques are then examined with the recent developments in the Monte Carlo integration by Markov Chain sampling. Two case studies are conducted with the aim of providing evidence of the applicability of Bayesian techniques. The first of these cases develops a procedure to test for seasonal and/or zero frequency unit roots in quarterly series. A new parameterisation is provided and the priors implemented are discussed and justified. The analysis relies on a Gibbs sampling scheme. The inferential technique used is the evaluation of posterior odds ratios. These ratios are defined as posterior expectations of functions of the parameters, and therefore can be consistently estimated The procedure is applied to some UK variables. The results are robust with respect to different prior distributions, and conflict with some conclusions reached by using classical asymptotic unit root tests. The second case study develops a Bayesian procedure to conduct inference in cointegrated systems. Inference regards the number of cointegrating relationships and their structural interpretation, and is based on the evaluation of highest posterior density confidence intervals. The procedure is applied to three VAR systems: Danish and Finnish money demands, and UK exchange rate data. Interesting results emerge, showing significant differences with their frequentist counterparts. All these results are robust with respect to different priors.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: HB Economic Theory