Long strange segments of a stochastic process. (Q1872448)

Long strange intervals in a linear stationary stochastic process with regularly varying tails are studied. Let \(X_n=\mu +\sum_{-\infty}^{\infty} \varphi_{n-j} Z_j\), \(n=1,2,\dots ,\) where \(\dots, Z_{-1}, Z_0, Z_1,\dots \) is a sequence of zero mean iid random variables and \(\mu\) is a constant. It is assumed that the tails of the distribution of \(Z=Z_1\) satisfy the regular variation condition \(P(| Z| >\lambda)=L(\lambda)\lambda ^{-\alpha}\), as \(\lambda\to\infty\), for some \(\alpha>1\), and \(L\) be a slowly varying function at infinity, and balance conditions \[ \lim_{\lambda\to \infty}{P(Z>\lambda)\over {P(| Z| >\lambda)}}=p, \quad \lim_{\lambda\to \infty}{P(Z<-\lambda)\over {P(| Z| >\lambda)}}=q, \] \(0<p=1-q\leq 1\). The behavior of the statistic \[ R_n(\theta)=\sup\left\{j-i:0\leq i < j \leq n, {X_{i+1}+\cdots + X_j\over {j-i}}> \theta \right\} \] is studied. \(R_n(\theta)\) can be considered as the greatest length of time interval when the system \(X_n\) runs under the nominal value. The main result of the paper is the limit theorem for the distribution of the vector \((R_n(\theta), M_n)\), where \(M_n=\max (X_1,\ldots, X_n)\). It is assumed that the sum of the modules of the coefficients \(\varphi_i \) is finite.