Conditions for convergence of random coefficient \(\mathrm{AR}(1)\) processes and perpetuities in higher dimensions (Q2448719)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Conditions for convergence of random coefficient \(\mathrm{AR}(1)\) processes and perpetuities in higher dimensions
scientific article

    Statements

    Conditions for convergence of random coefficient \(\mathrm{AR}(1)\) processes and perpetuities in higher dimensions (English)
    0 references
    0 references
    0 references
    5 May 2014
    0 references
    Several conditions for almost-sure convergence of the one-dimensional perpetuities \(Z_1+M_1Z_2+M_1M_2Z_3+\ldots\), where \((M_k,Z_k)_{k\in\mathbb{N}}\) are i.i.d. random elements of \(\mathbb{R}^2\), were given in [\textit{W. Vervaat}, Adv. Appl. Probab. 11, 750--783 (1979; Zbl 0417.60073)]. An ultimate criterion was obtained later by \textit{C. M. Goldie} and \textit{R. A. Maller} [Ann. Probab. 28, No. 3, 1195--1218 (2000; Zbl 1023.60037)] who, in particular, showed that a.s. convergence of the perpetuity entails \(\lim_{n\to\infty}\,M_1\cdot\ldots\cdot M_n=0\) a.s. whenever \(\operatorname{P}\{Z_1=0\}<1\) and \(\operatorname{P}\{M_1c+Z_1=c\}<1\) for all real \(c\). The present author is interested in a.s. convergence of multi-dimensional perpetuities, i.e., \((M_k,Z_k)_{k\in\mathbb{N}}\) are now i.i.d. random elements of \((\mathbb{R}^{d\times d},\mathbb{R}^d)\) for some integer \(d\geq 2\), where \(\mathbb{R}^{d\times d}\) is the space of \(d\times d\) matrices with real-valued entries, as well as distributional convergence of the corresponding Markov chains. He proves that, under the condition \(\lim_{n\to\infty}\,||M_1\cdot\ldots\cdot M_n||=0\) a.s. (condition \(C_0\)), where \(||\cdot||\) denotes the spectral norm, most of the criterion due to Goldie and Maller [loc. cit.] carries over to the multi-dimensional case with obvious modifications. What is missing is a counterpart of the condition \[ \int_{(1,\infty)}{\log x\over \operatorname{E}(\log^-|M_1|\wedge\log x) }\operatorname{P}\{|Z_1|\in \text{d}x\}<\infty. \] Also, the author demonstrates by a counterexample that a.s. convergence of multi-dimensional perpetuities does not imply \(C_0\). Further, it is shown that a.s. convergence of a multi-dimensional perpetuity does not entail distributional convergence of the corresponding Markov chain if \(C_0\) is not assumed. Some other refined counterexamples are given. The techniques used include (among others) an interesting application of the Kochen-Stone lemma and a non-trivial argument concerning the property to be a Cauchy sequence a.s.
    0 references

    Identifiers