Rates of convergence and approximations to the distribution of the maximum of chain-dependent sequence (Q1848520)

Let \(J_n\) be a finite state space Markov chain, let \(\{X_n\}\) be a sequence of r.v.s conditionally independent for a fixed \(J\) and let \(\Pr\{X_j\leq x\;|\;J_{j}=i\}=H_i(x)\). The authors analyze the accuracy of approximation of the maxima \(M_n=\max\{X_1,\dots,X_n\}\) distribution by \(\Lambda=\exp(-e^{-x})\) and consider the first-order corrections for this approximation. E.g., if \(H_i=1-h_i(x)e^{-x}\), where \(h_i\) are regularly varying functions with index \(\rho\),then \[ \Pr\{M_i\leq a_n x+b_n\}=\Lambda(x)+(x^2/2-2x)\Lambda'(x)\rho/b_n^2+o((\log n)^{-2}) \] for suitable \(a_n\) and \(b_n\). Results of simulations and applications to river flow data are presented.