Abstract
dc:descriptionThe study of time series has become a subject of interest last years especially for count data with heteroskedasticity. While there are numerous research studies with univariate models, in multivariate case the literature is limited. The reason for the above is that there are countless difficulties in the construction and analysis of multivariate counting processes. There are two general categories of time series models with different constructions leading to different ways of their theoretical properties analysis. While considering heteroskedastic time series models from the category ergodic properties can be studied more easier than the second one, methods for estimation are difficult to establish. This thesis deals with the construction of multivariate Integer Generalized Autoregressive Conditional Heteroskedastic(INGARCH) and Conditional Autoregressive Range(CARR) time series processes. The INGARCH(1,1) model has been also studied in multivariate case and implementations in 2 dimensions have been also presented recent years. The drawback was found at the restricted values of correlation coefficient boundaries depending on the way of model's construction. Based on a family of copulas a multivariate INGARCH(1,1) model and a CARR(1,1) model are studied considering interdependencies and self-dependencies respectively. According to model's complexity, its appropriateness and capability to study data with small sample sizes are examined and provided with simulations. H-steps ahead forecasting is considered by taking conditional expectation on volatilities and calculating marginal probability mass function. Firstly, considering univariate INGARCH models where volatilities are linearly or log-linearly expressed offering more flexibility on conditions of stationarity respectively, a Bayesian Trans-dimensional Markov Chain Monte Carlo is provided. At a second stage a new alternative family of Sarmanov distribution is also presented in order to ameliorate boundaries of correlation coefficient and comparison with the known Sarmanov families are graphically discussed. A multivariate INGARCH(1,1) process is studied based on this alternative Sarmanov distribution and an implementation with daily crash counts on three towns in Netherlands is presented. A multivariate Conditional Aytoregressive Range(CARR(1,1)) model assuming an exponential distribution and reconstructing correlation coefficients boundaries is discussed. The proposed model is illustrated on bivariate series from the fields of renewable sources.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsamtsakiri, Panagiota
- Τσαμτσακίρη, Παναγιώτα
Subjects
dc:subject × 11Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/56471