Athens University Economics and Business (AUEB)
Bayesian model determination and nonlinear threshold volatility models
Abstract
dc:descriptionThe purpose of this Thesis is to document an original contribution in the areas of model determination and volatility modeling. In this Thesis we focus on the development of efficient algorithms for Bayesian model determination using Markov Chain Monte Carlo (MCMC), which are also used to develop a family of nonlinear flexible models for volatility. We propose a new method for Bayesian model determination that incorporates several desirable characteristics, resulting in better mixing for the MCMC chain and more precise estimates of the posterior density. The new method is compared with various existing methods in an extensive simulation study, as well as more complex model selections problems based on linear regression, with both simulated and real data comprising of 300 to 1000 variables. The method seems to produce rather promising results, overperforming several other existing algorithms in most of the analyzed cases. Furthermore the method is applied to gene selection using logistic regression, with a famous dataset including 3226 genes. The problem lies in identifying the genes related to the presence of a specific form of breast cancer. We present a new class of flexible threshold models for volatility. In these models the variables included, as well as the number and location of the threshold points are estimated. To estimate these models we use the new method for Bayesian model determination, enriched with new move types, the use of which is validated through additional simulations. The new models are applied to estimate and predict the variance of the Euro-dollar exchange rate, using as exogenous variables a set of U.S. macroeconomic announcements. The results indicate that the threshold models can provide significantly better estimates and projections than other competitive conditional volatility models, while the most important macroeconomic announcements are identified. The threshold models are then generalised to the multivariate case. Under the proposed methodology, the estimation of the univariate variances is only required, as well as a rather small collection of regression coefficients.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2010
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Petralias, Athanassios
- Πετραλιάς, Αθανάσιος
Subjects
dc:subject × 20- Μπεϋζιανή
- Επιλογή υποδειγμάτων
- Υποδείγματα ορίων
- Διακύμανση
- Αλγόριθμοι
- Χρονολογικές σειρές
- Ισοτιμίες
- Επιλογή γονίδιων
- Bayesian
- Model selection
- Threshold models
- Volatility
- Algorithms
- Time series
- Exchange rates
- Gene selection
- Κοινωνικές Επιστήμες
- Άλλες Κοινωνικές Επιστήμες
- Social Sciences
- Other Social Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/23289