Athens University Economics and Business (AUEB)
Essays οn limit theorems for Martingale transforms with heavy-tailed innovations and the limit theory of the QMLE in conditionally heteroskedastic models
Abstract
dc:descriptionFinancial time series are known to exhibit fat tail behavior which is only partially captured by GARCH-type models. This has led to the consideration of heavy-tailed distributions for the squared innovation process. The purpose of this thesis is to derive limit theorems with mixed stable limits and, using the latter, to examine the asymptotic theory of the Gaussian quasi-maximum likelihood estimator in a general class of conditionally heteroskedastic models relaxing the usual assumption of finite fourth moments on the innovation process. The first chapter concerns the limit theory of the QMLE for the nonlinear GQARCH(1,1) model under the setting of strict stationarity and ergodicity. Using the stochastic recurrence equation (SRE) approach we establish conditions for stationarity-ergodicity and conjecture that they are identical with the GARCH(1,1) model while proving an analogous argument for the QARCH(1) model. We proceed to determine the limit behavior of the QMLE allowing the squared innovation process to lie in the normal domain of attraction of an \alpha-stable distribution with index of stability \alpha\in(1,2] and the true parameter vector to lie on the boundary of the parameter space resulting in weak limits which are thus allowed to be non-normal. In the second chapter we extend an existing result in the literature, which establishes asymptotic normality of the QMLE in the non-stationary ARCH(1) model, by allowing the innovation process not to possess finite fourth moments. Specifically we assume that the squared innovation process lies in the domain (not necessarily normal) of attraction of an \alpha-stable distribution with \alpha\in(0,2]. When \alpha\in[1,2] we obtain \alpha-stable limits with the same \alpha, while in the case where \alpha\in(0,1) we derive inconsistency and non-tightness for the QMLE. In chapter 3 we provide a CLT for martingale transforms which holds even for sequences with infinite variance in which case we obtain normal limits but with rate of convergence different than the usual \sqrt{n}. Specifically, the rate will depend on the rate of divergence of the second truncated moments. Interestingly enough the result obtains by verifying a modified version of Lindeberg's condition under mild conditions and the rate of convergence can be easily estimated. In chapter 4 we provide a limit theorem with mixed stable limits that is general enough to be applicable in a variety of cases, including those examined in the previous chapters. We provide a general framework for stationary and ergodic conditionally heteroskedastic processes for the QMLE limit theory using stochastic recurrence equation methods. Then we derive the limit theory of the QMLE when \alpha\in[1,2), and inconsistency and non-tightness when \alpha\in(0,1). Furthermore, we provide an example where the stationarity framework can be relaxed by deriving the limit theory of the QMLE for the non-stationary GARCH(1,1). Finally, we determine the asymptotic behavior of the usual Wald statistic and obtain asymptotic exactness and consistency of the relevant testing procedure based on subsampling methods. Then we evaluate the performance of the latter numerically.
Degree
thesis:*- Grantor dc:publisher
- Athens University Economics and Business (AUEB)
- Year dc:date
- 2015
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Louka, Alexandros
- Λουκά, Αλέξανδρος
Subjects
dc:subject × 11Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/37088