Remarks on the speed of convergence of mixing coefficients and applications (Q2435775): Difference between revisions

Convergence rates of mixing coefficients for Markov chains generated by copulas are studied. New conditions are given for a copula-based Markov chain to be \(\rho\)-mixing, exponentially \(\rho\)-mixing and geometrically ergodic. The copula associated to the Metropolis-Hastings algorithm is derived. Then, conditions on such copula are given so that the Metropolis-Hastings algorithm generates an exponentially \(\rho\)-mixing Markov chain. Also, assumptions for the generated Markov chain to be geometrically ergodic and for it to be uniformly mixing are given. Another type of results presented here has to do with mixing coefficients for convex combinations of copula-based Markov chains, i.e., given a set of copulas, under what conditions any convex combination of such set generates a geometrically ergodic Markov chain. From that, we learn that the Mardia family and the Fréchet family of copulas generate \(\rho\)-mixing and ergodic reversible Markov chains, if a slight condition is satisfied. Hence, the question whether convex combinations of geometrically ergodic reversible Markov chains are geometrically ergodic is settled.