Spectral properties of superpositions of Ornstein-Uhlenbeck type processes

DOI10.1007/S11009-005-4521-0MaRDI QIDQ2583517zbMATH OpenOpenAlexFDO

Authors Ole E. Barndorff-Nielsen, Nikolai N. Leonenko

Publication date 17 January 2006

Published in Methodology and Computing in Applied Probability (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1007/s11009-005-4521-0

correlation function long-range dependence Mittag-Leffler function normal inverse Gaussian distribution stationary processes

Mathematics Subject Classification ID

Infinitely divisible distributions; stable distributions (60E07) Stationary stochastic processes (60G10) Self-similar stochastic processes (60G18) Time series, auto-correlation, regression, etc. in statistics (GARCH) (62M10)

Review text

The introducion presents a short review of investigations and possible applications of the results in this field. Further, much attention is paid to a number of known results that serve as the background for calculations. Several new examples of continuous time strictly stationary processes are obtained whose autocorrelation functions and spectra are of simple form, the marginals are of special form. The processes exhibit long-range dependence. Let \(y\) be a self-decomposable variable with differentiable Lévy density, let \(z(s)\) be the Lévy process generalized by \(y,\) i.e., \(z(1) \overset{L}{=}y.\) There is a certain stochastic differential equation having a stationary solution for any \(\lambda >0\), \[ y(u) =e^{-\lambda | u| }y(0) +\int_{0}^{u}e^{-\lambda | u-s|}\,dz_{1}(\lambda s) ,\quad u\in R, \] where \(z_{1}(s)\) is the background driving Lévy process corresponding to the r.v. \(y=z(1).\) When \(y(u)\) is square integrable with \(Ey(u) =0,\) it has the correlation function \(z(u) =\exp \{ -\lambda | u| \} .\) The process \(y(u)\) is called the Ornstein-Uhlenbeck type process or OU type, for short. When the r.v. \(y\) has the normal inverse Gaussian distribution (NIG\(( \alpha ,\beta ,\mu ,\delta)\)), the OU type processes are discussed in Example 1. Next examples are based on superpositions of the OU type processes. In a discrete type of superposition, let there exist a sequence \(\{ y^{( k) }( u) \} \) of OU type processes such that, for all \(k\) and \(u\in R\), the marginal distribution of \(y^{( k) }( u) \) is NIG\(( \alpha ,\beta ,\delta _{k},0) \) with \[ \delta _{k}=\frac{1}{k^{1+2( 1-H) }},\quad k=1,2,\ldots, \quad H\in ( 0,1), \] and its correlation function \(r( u) =e^{-\lambda | u| }\), \(u\in R.\) Let \(\delta =\sum_{k=1}^{\infty }\delta _{k}.\) Then the process \[ y( u) =\sum_{k=1}^{\infty }y^{( k) }( k_{n}^{-1}) ,\quad u\in R, \] is stationary and well defined as an \(L_{2}\)-limit and its marginal is NIG\( ( \alpha ,\beta ,\delta _{k},0).\) Its correlation function \(r(\tau)\) together with its asymptotics are presented as \(| \tau | \to \infty,\) as well as the normalized spectral density with asymptotics for \(| \tau | \to 0\) and \(| \tau | \to \infty.\) Note that instead of NIG\(( \alpha ,\beta ,\delta _{k},0) \) some other distributions can be used. The continuous type of superpositions of OU processes is considered via an independently scattered random measure \(z=\{ z( A) ,A\in S\}\) on \(T\subseteq R^{d}.\) Here \(T=R\times R_{+}\) with points \(w=(s,\xi)\) and \(z\) is an independently scattered measure on \((T,B)\) with Lévy characteristics \(0;0;\overset{} {q}(A,dx) =M(A) W(dx)\) for some Lévy measure \(W\) on \(R\) and the finite measure \(M\) on \(T\) such that \(M(dw) =ds\nu (d\xi)\) with a probability measure \(\nu\) on \(R_{+}.\) For \(u\in R\) and \(B\in \Re (R_{+}),\) it is possible to define \[ y(u,B) =\int_{B}e^{-\xi u}\int_{-}^{\xi u}e^{s}z(ds,d\xi). \] Then the process \(y(u) =y(u,R_{+}),\) is a well-defined, infinitely divisible and strictly stationary process. Such a process is called a sup-OU process. If the random process \(y\) is square integrable, then its correlation function is \[ r(\tau)=\int_{0}^{\infty}e^{-\tau \xi }\nu (d\xi) ,\quad \tau >0, \] and the spectral density \[ f(\lambda) =\frac{1}{\pi }\int_{0}^{\infty }\frac{\xi }{ \xi ^{2}+\lambda ^{2}}\nu (d\xi). \] The authors obtain correlation functions and spectral densities in the case where \(\nu \) is the Linnik and generalized Linnik distribution, and if \(\nu \) is the Mittag-Leffler, or generalized Mittag-Leffler, or two-parameter Mittag-Leffler distribution. The long-range dependence and the possibility of Mittag-Leffler decay in the autocorrelation function are discussed. The authors summarize the results of the correlation functions and spectral densities in Table 1 and of their asymptotics in Table 2.

Recommendations

Cites work

Cited in

(54)

This page was built for publication: Spectral properties of superpositions of Ornstein-Uhlenbeck type processes

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2583517)