Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.638325
Title: Some aspects of parameter estimation in time series transfer function models
Author: Ngadiman, S. B.
Awarding Body: University College of Swansea
Current Institution: Swansea University
Date of Award: 1995
Availability of Full Text:
Access through EThOS:
Abstract:
In this thesis several general aspects of inferences on parameters in time series transfer function models are considered. Chapter 1 is a general discussion on fitting transfer functions to data, and highlights several issues for further examination. These include model representation, parameter estimation and model misspecification. Two general representations of transfer function model are considered: a linear form, Model (I), and a rational form, Model (II). The non-uniqueness in representation is discussed, and the resultant bias in estimates of parameters in certain models is considered in Chapter 2. The log-likelihood function is developed for some linear transfer function models in Chapter 3, which also contains a discussion of asymptotic behaviour of estimates. Chapter 4 completes the discussion on inferences of parameters by considering the estimation of noise variance through a profile log-likelihood. The same framework is then used to study estimates of parameters in a misspecified transfer function model; this discussion is presented in Chapter 5 and Chapter 6. A further examination on the relationship between true model parameters from a linear model of order (0,2,0,0) and the estimates of parameters in a fitted model of order (0,0,0,1) is also given. Chapter 7 reviews forecasting methods and the behaviour of forecast mean square error in misspecified models. Throughout, our discussions include extensive numerical examples; the univariate case is used for comparison where appropriate.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.638325  DOI: Not available
Share: