Extremes of integer-valued moving average models with regularly varying tails (Q1887252)

Let \(\{X_{n}\}\), \(n\in N\), be a stationary sequence of non-negative integer-valued random variables defined by \(X_{n}=\sum_{i=-\infty}^{\infty}\beta_{i}\circ Z_{n-i}\), \(\beta_{i}\in[0,1]\), where \(\{Z_{n}\}\) is an independent an identically distributed sequence of non-negative integer-valued random variables and \(\circ\) denotes binomial thinning defined as \(\beta\circ Z=\sum_{i=1}^{Z}B_{i}(\beta)\), where \(\{B_{i}(\beta)\}\) is an i.i.d. sequence of Bernoulli random variables with \(P\{B_{i}(\beta)=1\}=\beta\). All thinning operations in the definition of \(X_{n}\) are independent for each \(n\). The author proves that if the distribution of the innovations \(Z_{n}\) has a regularly varying tail verifying \[ n(1-F_{Z}(u_{n}))\to \tau \left(\sum_{i=-\infty}^{\infty} \beta_{i}^{\alpha}\right)^{- 1},\quad \tau>0, \text{ as }n\to\infty, \] for some sequence \(\{u_{n}\}\), then the distribution of \(X_{n}\) satisfies \(n(1-F_{X}(u_{n}))\to\tau\), as \(n\to\infty\), and the sequence has the extremal index \(\theta=\max_{-\infty<i<\infty}(\beta_{i}^{\alpha}) (\sum_{i=-\infty}^{\infty} \beta_{i}^{\alpha})^{-1}\). A simulation study illustrating the obtained results is presented.