A note on the comparison of stationary laws of Markov processes (Q805075)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | A note on the comparison of stationary laws of Markov processes |
scientific article |
Statements
A note on the comparison of stationary laws of Markov processes (English)
0 references
1991
0 references
Let \(\{X_ n\}\), \(\{Y_ n\}\) be two Markov chains on \({\mathfrak R}^ k\) generated by the autoregression functions \(T_ 1\) and \(T_ 2\), respectively: \[ X_{n+1}=T_ 1(X_ n)+Z_ n,\quad Y_{n+1}=T_ 2(Y_ n)+Z_ n, \] where the \(Z_ n's\) are i.i.d. random variables independent of the X- and Y-processes. In addition, it is assumed that \(\{X_ n\}\) and \(\{Y_ n\}\) admit unique stationary distributions \(\mu_ X\) and \(\mu_ Y\). (This condition is fulfilled, e.g., if the T's are contractions with contractivity q strictly less than 1 and the Z's have an absolutely continuous distribution supported by all of \({\mathfrak R}^ k.)\) The paper introduces a partial order of probability distributions measuring the degree of concentration around zero. The author studies the influence of the autoregression function T and the distribution of the noise term \(Z_ n\) on this degree of concentration, the intuitive idea being that the more T is contracting, the more concentrated is the stationary distribution around the fixed point of T (the origin). It is shown that \(\mu_ X\) is more concentrated around the origin than \(\mu_ Y\) if \(T_ 1\) is more contracting than \(T_ 2\) in a rather strong sense and, moreover, the density of the noise distribution exhibits a certain ``spherical'' symmetry. Elementary examples indicate, see, e.g., the study of the corresponding one-dimensional problem by the reviewer [J. Time Ser. Anal. 7, 205-211 (1986; Zbl 0623.60086)], that the strong assumptions made cannot be relaxed very much.
0 references
conparison theorem
0 references
stochastic ordering
0 references
Markov chains
0 references
autoregression functions
0 references
unique stationary distributions
0 references
degree of concentration
0 references
0 references