Nonstandard limit theorem for infinite variance functionals (Q2482288)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Nonstandard limit theorem for infinite variance functionals |
scientific article |
Statements
Nonstandard limit theorem for infinite variance functionals (English)
0 references
16 April 2008
0 references
Let \(\{X_i\}\) be a stationary Gaussian sequence defined by \[ X_i = \sum_{j=0}^\infty b_j \xi_{i-j}, \] where the \(\xi_i\) are independent standard Gaussian random variables and \(b_j = j^{H - 3/2} L_1(j)\), with \(1/2 < H < 1\), \(L_1(j)\) slowly varying and \(\sum_{j=0}^\infty b_j^2 = 1\). Then \(\{X_i\}\) are long-range dependent with Hurst index \(H\). If \(f\) is a function such that \(E \big[f(X_1)^2\big] < \infty\), then the weak limits of normalized sums of the form \(\sum_{i=1}^{\lfloor nt \rfloor} \big[f(X_i) - Ef(X_i)\big]\) have been investigated by \textit{R. L. Dobrushin} and \textit{P. Major} [Z. Wahrscheinlichkeitstheor. Verw. Geb. 50, No. 1, 27--52 (1979; Zbl 0397.60034)], \textit{M. S. Taqqu} [Z. Wahrscheinlichkeitstheorie Verw. Geb. 50, 53--83 (1979; Zbl 0397.60028)] and \textit{P. Breuer} and \textit{P. Major} [J. Multivariate Anal. 13, 425--441 (1983; Zbl 0518.60023)]. In this paper under review, the authors consider the limits of normalized sums when \(f(X_1)\) has power-tailed distribution with index \(0 < \alpha < 2\): \[ P(f(X_1) > x) \sim \frac{1 + \beta} 2 L_2(x) x^{-\alpha}, \] \[ P(f(X_1) < - x) \sim \frac{1 - \beta} 2 L_2(x) x^{-\alpha} \] as \( x \to \infty\), where \(L_2\) is slowly varying and \(\beta \in [-1, 1]\) is a constant. Assume that \(E\big| f(X_1)\big| ^p < \infty\) for some \(p > 1 \) and let \(\kappa\) be the (extended) Hermite rank of \(f\). The authors show, among other things, that (i) If the long-range dependence is strong enough (i.e., \(1 - \kappa (1 - H) > 1/\alpha\)), then the limit of normalized sums is (up to a constant) the \(\kappa\)th Hermite process. (ii) If \(1 - \kappa (1 - H) < 1/\alpha\), then the limit is \(\alpha\)-stable Lévy motion. (iii) If \(1 - \kappa (1 - H) = 1/\alpha\) and some additional conditions are satisfied, then the limit is a sum of a Hermite process and an \(\alpha\)-stable Lévy motion.
0 references
Fractional Brownian motion
0 references
long-range dependence
0 references
stable law
0 references
hypercontractivity
0 references
0 references