The Optimal Linear Combination of Control Variates in the Presence of Bias.
Abstract
The method of control variates has been extensively studied as a technique for obtaining variance reductions for complex simulations. The method basically requires that the practitioner be able to identify processes for which the asymptotic mean is known; the knowledge of those asymptotic means is then used to obtain a variance reduction. The authors' goal is to study a specific aspect of the small-sample theory for control variates. Their particular interest focuses on the loss of efficiency incurred when only the asymptotic mean is known, as opposed to the true (small-sample) mean. The results obtained have implications for the application of control variates to the steady-state estimation problem. Specifically, in many steady-state simulations, only the asymptotic means of the control variates are known. The results obtained here complement other small-sample studies on control variates in which the focus is on the degradation in performance caused by estimation of the optimal control coefficients. Our methods can also be used to study small-sample properties of the method of multiple estimates.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1986
- Accession Number
- ADA169111
Entities
People
- Donald Iglehart
- Peter W. Glynn
Organizations
- Stanford University