Estimation in conditionally heteroscedatic time series models. (Q2386889): Difference between revisions

The book deals with conditionally heteroscedastic time series models. It covers classical and new topics of parameter estimation in such models. The author studies a general conditionally heteroscedastic time series model of the form \[ X_t=\sigma_t Z_t, \quad\sigma_t^2=g_\theta (X_{t-1},\dots,X_{t-p},\sigma_{t-1}^2,\dots,\sigma_{t-q}^2), \quad t\in Z.\tag{1} \] He starts with basic concepts and a survey of known financial time series models, and their properties and probabilistic tools for their investigation. Then a short overview on the estimation in ARMA models, quasi maximum likelihood (QMLE) and Whittle estimators for GARCH processes, and asymptotic properties, introduces the readers to the problems and methods considered in the following chapters. In the next chapter, the author develops a unifying theory for QMLE in model (1) which is based on techniques of stochastic recurrence equations. Applications to AGARCH and EGARCH time series are demonstrated. The maximum likelihood estimator (MLE) in the model (1) is investigated in chapter 6. Consistency and asymptotic normality of the MLE are established. A case of Student innovations \(Z_t\) is considered. In this part of the book, problems related to misspecification of the innovations distribution are discussed. In the next chapter, QMLE in model (1) with heavy-tailed innovations is studied. The last chapter is devoted to investigation of the properties of the Whittle estimator in a heavy-tailed GARCH(1,1) model. There are a lot of various examples and remarks which clarify the presented general results. Some numerical examples and simulations are given. Detailed discussions and comparisons with known results are presented in each chapter.