Extremal indices, geometric ergodicity of Markov chains and MCMC (Q2463676)

The extremal index \(\vartheta\) of a stationary random sequence indicates dependence of the extremal events (e.g. the exceedances of a high level): \(\vartheta=1\) corresponds to the independence, \(\vartheta=0\) represents the strong clustering of extremal events. Connections between \(\vartheta\) and geometrical ergodicity of Markov chains are investigated with special attention to the chains generated by a Metropolis algorithm (MA). It is shown that if a stationary distribution of MA has a long tail, then \(\vartheta=0\). Geometrically ergodic (GE) chains under some mild condition have \(\vartheta>0\). Explicit expressions for \(\vartheta\) of GE MA are given.