An important parameter of the regenerative method of simulation output
analysis is the choice of return state used for blocking observations
. Computational experience has shown that the statistical properties o
f estimators based on different regeneration points can vary widely. I
n this paper we study the limiting joint distribution of the normalize
d regenerative point and standard-deviation estimators for general sta
te-space Markov chains. The asymptotic covariance between the point an
d standard-derivation estimators is shown to be the same for all retur
n states, and the quantity is related to the normalized skewness of th
e partial sums of the chain. Since the covariance is constant, it foll
ows that the choice of return state that minimizes the asymptotic vari
ance of the standard-deviation estimator will maximize the correlation
between the point and standard-deviation estimator. Consideration of
asymptotic variance and covariance alone suggests that confidence inte
rvals have the best coverage probabilities if the asymptotic variance
of the standard-deviation estimator is minimized (thereby maximizing t
he correlation).