Extremes of Volterra series expansions with heavy-tailed innovations (Q2573537)

There is no unifying theory that is applicable to all nonlinear systems and consequently the study of such systems has to be restricted to special classes of nonlinear models. Volterra series expansions or polynomial systems are such class. It is well known that every stationary time series may be approximated, in a certain sense, by two-sided Volterra series of the form: \[ X_k =\sum_{i=1}^\infty\left(\sum_{| j_1| \leq \infty,\dots,| j_i| \leq\infty} h_{j_1,j_2,\dots,j_i} \prod_{r=1}^i Z_{k-j_r}\right),\quad k=1,2,\dots, \] where \(\{Z_t\}\) is a sequence of independent and identically distributed innovations and \(\{h_{j_1}\}\), \(\{h_{j_1,j_2}\}\),... are such that the sum converges to a well defined random variable. This equation is known as the discrete time Volterra series expansion and the sequences \(\{h_{j_1}\}\), \(\{h_{j_1,j_2}\},\dots\) are called the kernels of Volterra series. Volterra series expansions are known to be the most general nonlinear representation for any stationary sequence. The first use of Volterra series expansions in nonlinear system theory was due to Wiener, who was concerned with the analysis of physical systems. Volterra series expansions have also found significant applications in signal processing such as image, video and speech processing where the nonlinearities are mild enough to allow approximations by low order Volterra polynomial expansions; see for example \textit{V. J. Mathews} and \textit{G. L. Sicuranza} [``Polynomial signal processing'' (Wiley, New York, 2000)] and references therein. In view of these facts, the study of the extremal properties of finite-order Volterra series expansions would be highly valuable in understanding the extremal behavior of nonlinear processes as well as understanding of order identification and adequacy of Volterra series when used as models in signal processing. In fact, such extremal properties may suggest a way of finding the order of a finite Volterra expansion which is consistent with the nonlinearities of the observed process. The paper is organized as follows: Section 2 deals with the tail behavior of \(X_k\), when the sequence of innovations \(\{Z_k\}\) is heavy-tailed. Section 3 provides a detailed point process analysis of asymptotic properties of the \(m\)th-order Volterra series expansions and gives some complementary extreme value results. In particular, the authors deduce the maximum limiting distribution and the extremal index, which has an informal interpretation as the reciprocal of the average size of clusters of high values. These sections depend on the results given by \textit{S. Resnick} and \textit{E. van den Berg} [Commun. Stat., Stochastic Models 16, 233--258 (2000; Zbl 0955.60028)]. In Section 4 some examples of bilinear processes are given and the known results on extremes of such processes are compared with the corresponding adequate finite-order Volterra series approximations. Finally, a simulation study is devised to assess the performance of the suggested method.