Comparison on five estimation approaches of intensity for a queueing system with short run (Q638038)

Confidence intervals construction is considered for \(\rho= E Y/E X\) by i.i.d. samples \(X_1\),\dots, \(X_n\) and \(Y_1\),\dots, \(Y_m\). The authors discuss the asymptotic plug-in interval based on CLT and four bootstrap-type procedures: standard, Bayesian, percentile and accelerated bias corrected bootstrap confidence intervals. Performance of the algorithms is assessed via simulations. Special attention is paid to the case when \(X\) and \(Y\) represent the interarrival and service times of a queueing system (and \(\rho\) is the utilization factor or the intensity of the system). The authors' conclusion is that the Bayesian bootstrap outperforms the other four approaches in terms of relative coverage.