Initializing for bias reduction: some analytical considerations

  • Authors:

  • Joseph R. Murray;W. David Kelton

  • Affiliations:

  • Dept. of Industrial and Operations Engineering, The University of Michigan, Ann Arbor, MI;Dept. of Management Sciences, University of Minnesota, Minneapolis, MN

  • Venue:
  • WSC '88 Proceedings of the 20th conference on Winter simulation
  • Year:
  • 1988

Quantified Score

Hi-index 0.00

Visualization

Abstract

We are concerned with simulation studies where the simple method of independent replications and classical statistical techniques are used to construct estimates (point and interval) for a steady-state parameter of interest. This paper describes the results of an analytical study on the effectiveness of stochastic initialization, where the initial state of each replication is picked randomly from an initial state distribution. The initial-state distribution is constructed from data observed during a “pilot phase”. Using an AR(1) process, and assuming a fixed simulation budget, we show that stochastic initialization is effective in reducing bias in the point estimate and increasing coverage of the interval estimate without unduly increasing variance or mean square error.