Abstract This thesis considers some linear and nonlinear time series models. In the linear case, the analysis of a large number of short time series generated by a first order autoregressive type model is considered. The conditional and exact maximum likelihood procedures are developed to estimate parameters. Simulation results are presented and compare the bias and the mean square errors of the parameter estimates. In Chapter 3, five important nonlinear models are considered and their time series properties are discussed. The estimating function approach for nonlinear models is developed in detail in Chapter 4 and examples are added to illustrate the theory. A simulation study is carried out to examine the finite sample behavior of these proposed estimates based on the estimating functions.

Abstract This thesis considers some linear and nonlinear time series models. In the linear case, the analysis of a large number of short time series generated by a first order autoregressive type model is considered. The conditional and exact maximum likelihood procedures are developed to estimate parameters. Simulation results are presented and compare the bias and the mean square errors of the parameter estimates. In Chapter 3, five important nonlinear models are considered and their time series properties are discussed. The estimating function approach for nonlinear models is developed in detail in Chapter 4 and examples are added to illustrate the theory. A simulation study is carried out to examine the finite sample behavior of these proposed estimates based on the estimating functions.

The purpose of the present study is to investigate the predictive validity of the BUEPE through the use of the three-parameter IRT model&amp / #8217 / s ability estimates. The study made use of the BUEPE September 2000 data which included the responses of 699 students. The predictive validity was established by using the departmental English courses (DEC) passing grades of a total number of 371 students. As for the prerequisite analysis the best fitted model of IRT was determined by first, checking the assumptions of IRT / second, by analyzing the invariance of ability parameters and item parameters and thirdly, by interpreting the chi-square statistics. After the prerequisite analyses, the best fitted model&amp / #8217 / s estimates were correlated with DEC passing grades to investigate the predictive power of BUEPE on DEC passing grades. The findings indicated that the minimal guessing assumption of the one- and two-parameter models was not met. In addition, the chi-square statistics indicated a better fit to the three-parameter model. Therefore, it was concluded that the best fitted model was the three-parameter model. The findings of the predictive validity analyses revealed that the best predictors for DEC passing grades were the three-parameter model ability estimates. The second best predictor was the ability estimates obtained from sixty high information items. In the third place BUEPE total scores and the total scores obtained from sixty high information items followed with nearly the same correlation coefficients. Among the three sub-tests, the reading sub-test was found to be the best predictor of DEC passing grades.

This thesis is devoted to the models that are suitable for modelling spatial data. For this purpose, random fields with finite index set are used. Based on the neighbourhood relationship a spatial weight matrix is introduced which describes spatial dependencies. A recognition and testing of spatial dependence is mentioned and it is applied for macroeconomic indicators in the Czech Republic. Spatial models originated from generalization of usual time series models are subsequently combined with linear regression models. The parameter estimators are derived for selected models by three different methods. These methods are ordinary least squares, maximum likelihood and method of moments. Theoretical asymptotic results are supplemented by a simulation study that examines the performance of estimators for finite sample size. Finally, a short illustration on real data is demonstrated. Powered by TCPDF (www.tcpdf.org)

The aim of this thesis is a comprehensive description of the properties of a nonlinear least squares estimator for a nonlinear regression model with normally distributed errors and thorough development of various methods for constructing confidence regions and confidence intervals for the parameters of the nonlinear model. Due to the fact that, unlike the case of linear models, there is no easy way to construct an exact confidence region for the parameters, most of these methods are only approximate. A short simulation study comparing observed coverage of various confidence regions and confidence intervals for models with different curvatures and sample sizes is also included. In case of negligible intrinsic curvature the use of likelihood-ratio confidence regions seems the most appropriate.

The research attempts to resolve which method of estimation is the most consistent for the parameters of the earthquake model, and how these different methods of estimation, as well as other changes, in the earthquake model parameters affect the damage estimates for a specific area. The research also investigates different methods of parameter estimation in the context of the log-linear relationship characterised by the Gutenberg-Richter relation. Traditional methods are compared to those methods that take uncertainty in the underlying data into account. Alternative methods based on Bayesian statistics are investigated briefly. The efficiency of the feasible methods is investigated by comparing the results for a large number of synthetic earthquake catalogues for which the parameters are known and errors have been incorporated into each observation. In the second part of the study, the effects of changes in key parameters of the earthquake model on damage estimates are investigated. This includes an investigation of the different methods of estimation and their effect on the damage estimates. It is found that parameter estimates are affected by observation errors. If errors are not included in the method of estimation, the estimate is subject to bias. The nature of the errors determines the level of bias. It is concluded that uncertainty in the data used in earthquake parameter estimates is largely a function of the quality of the data that is available. The inaccuracy of parameter estimates depends on the nature of the errors that are present in the data. In turn, the nature of the errors in an earthquake catalogue depends on the method of compilation of the catalogue and can vary from being negligible, for single source catalogues for an area with a sophisticated seismograph network, to fairly impactful, for historical earthquake catalogues that predate seismograph networks. Probabilistic seismic risk assessment is used as a catastrophe modelling tool to circumvent the problem of scarce loss data in areas of low seismicity and is applied in this study for the greater Cape Town region in South Africa. The results of the risk assessment demonstrate that seemingly small changes in underlying earthquake parameters as a result of the incorporation of errors can lead to significant changes in loss estimates for buildings in an area of low seismicity. / Dissertation (MSc)--University of Pretoria, 2014. / Insurance and Actuarial Science / MSc / Unrestricted