Joint extremal behavior of hidden and observable time series with applications to GARCH processes (Q2340041)

The authors study the extremal behavior of a class of generalized hidden Markov models \(\left( X_{t},Y_{t}\right) _{t\in\mathbb{Z}}\), where \[ X_{t}=\Psi\left( Y_{t-s},\epsilon_{t-s},\dots{},\epsilon_{t}\right) ,\quad t\in\mathbb{Z}\text{,} \] for some measurable mapping \(\Psi:\mathbb{R\times S}^{s+1}\rightarrow \mathbb{R}\) with \(s\geq0\)\ and \[ Y_{t}=\Phi\left( Y_{t-1},\epsilon_{t-1}\right) ,\quad t\in\mathbb{Z}\text{,} \] for some measurable mapping \(\Phi:\mathbb{R\times S}\rightarrow\mathbb{R}\). Let \(\left( \epsilon_{t}\right) _{t\in\mathbb{Z}}\)\ be a sequence of i.i.d.\ innovations taking values in a Polish space \(\mathbb{S}\). Suppose that \(\left( \epsilon_{t}\right) _{t\geq v}\) is independent of \(\left( Y_{t}\right) _{t\leq v}\) for all \(v\in\mathbb{Z}\). Then, \(\left( Y_{t}\right) _{t\in\mathbb{Z}}\) is a Markov process. This class of processes contains, e.g., the broad class of GARCH(1,1) processes. In this paper, the authors focus on the limit distribution of \[ \mathcal{L}\left( \frac{Y_{m}}{x},\dots{},\frac{Y_{n}}{x}\Big |\left| X_{0}\right| >x\right) ,\quad-\infty<n\leq m<\infty, \] as \(x\rightarrow\infty\).\ They give sufficient conditions for the existence of this limit and characterize its special structure. The authors also analyze connections of their results with multivariate regular variation of the time series \(\left( X_{t},Y_{t}\right) _{t\in\mathbb{Z}}\). The theoretical results are applied to financial asset returns by the asymmetric GARCH(1,1) model. They allow for a simple representation of the tail chains in this case and are used for Monte Carlo evaluations of some extremal characteristics.