Likelihood estimation of the extremal index (Q111096): Difference between revisions

Let \(X_1\),\dots,\(X_n\) be a stationary sequence. The author considers estimation of the extremal index \(\vartheta\) of the sequence based on the approximated likelihood of the gaps \(S_i\) between the exceedances over a high threshold \(u_n\). More precisely, let \(N\) be the number of the exceedances and let \(j_i\), \(i=1\),\dots,\(N\), be their locations. Then \(S_i=j_{i+1}-j_i-1\). One version of the estimators considered in the paper is \[ \hat\vartheta={T+N-1+N_c-[(T+N-1+N_c)^2-8N_cT]^{1/2}\over 2T}, \] where \(T=\sum_{i=1}^{N-1} qS_i\), \(N_c=\sum_{i=1}^{N-1} 1_{S_i\not=0}\), and \(q\) is an estimate for \(P(X_i>u_n)\). The applicability of \(\hat\vartheta\) to the estimation of the local extremal index in nonstationary data is investigated via simulations. An application to the Central England temperature data from 1772 through 2004 is considered.