On the extremal behaviour of generalized periodic sub-sampled moving average models with regularly varying tails (Q2488449): Difference between revisions

Let \(Z_i\) be i.i.d. with regularly varying tails, and let \(X_k\) be a stationary sequence of the form \(X_k=\sum_{j=-\infty}^{+\infty}\beta_j*Z_{k-j}\) (here \(*\) means multiplication for real valued \(Z_i\) and binomial thinning for nonnegative integer-valued \(Z_i\)). The observed data \( Y_k\) are sub-sampled from \(X_k\) by the rule \(Y_k=X_{g(k)}\), where \(g(k)\) is some strictly increasing sequence with a periodic pattern of the form \(g(k+I)=g(k)+M\) for some fixed \(I\), \(M\) and all \(k\). The tail behavior of \(\underline{M}_n=\max_{1\leq g(k)\leq n} Y_k\) is investigated. E.g., if \(Z\) belongs to the domain of max attraction of the \(\alpha\)-Fréchet distribution, \(a_n\sim(\sum_{j=-\infty}^{+\infty}\beta_j^\alpha)F^{-1}(1-n^{-1})\), then \[ \lim_{n\to\infty}{\mathbf P}\{\underline{M}_n\leq a_n x\}=e^{-M^{-1}\vartheta_Mx^{-\alpha}}, \] \[ \text{where}\quad \vartheta_M={\sum_{i=0}^M\vee_{-\infty<t<\infty}\{\beta^\alpha_{\bar g(t)+i}\}}, \] and \(\bar g\) is the extension of \(g\) on \({\mathbb Z}\).