Multivariate estimation and variance reduction in terminating and steady-state simulation

ABSTRACT

Research on the analysis of steady-state simulation experiments has concentrated on mitigating the effects of initial-condition bias and estimating the variance of the simulation point estimator, usually a sample mean. There has been little research on improving the precision of point estimators through variance reduction, especially in multivariate estimation problems. In fact, multivariate estimation procedures are rarely used in simulation output analysis. We consider applying the non-overlaping batch means output analysis method in conjunction with the control-variates variance reduction technique to estimate a multivariate mean vector. The effect of the number of batches and the number of control variates on the multivariate point and region estimators and the univariate point and interval estimators are considered. Our results have implications for terminating simulations as well.

REFERENCES

• 1: Rowena Añonuevo , Barry L. Nelson, Automated estimation and variance reduction via control variates for infinite-horizon simulations, Computers and Operations Research, v.15 n.5, p.447-456, November 1988 [doi>10.1016/0305-0548(88)90061-5]
• 2: Bauer, K.W. (1987). Control variate selection for multiresponse simulation. Unpublished Ph.D. dissertation. School of Ind~=strial Engineering, Purdue University.
• 3: Der-Fa R. Chen , Andrew F. Seila, Multivariate inference in stationary simulation using batch means, Proceedings of the 19th conference on Winter simulation, p.302-304, December 14-16, 1987, Atlanta, Georgia, United States [doi>10.1145/318371.318436]
• 4: Nelson. B.L. (1986). Batch size effects on the efficiency of control vaxiates in simulation. Working Paper Series No. 1986- 001. Department of Industrial and Systems Engineering, The Ohio State University'.
• 5: Nelson. B.L. (1987). A perspective on variance reduction in dynamic simulation experiments. Communications in Statistic, B 16,385-426.
• 6: Lee W. Shruben, Control of initialization bias in multivariate simulation response, Communications of the ACM, v.24 n.4, p.246-252, April 1981 [doi>10.1145/358598.358634]
• 7: Venkatr,~m~m, S. and Wilson, J.R. (1986). The efficiency of control variates in rnultiresponse simulation. O~e,'a~ian~ Research L~'.~ters .ti, 37-42.
• 8: Wil'~on, ,{.R. (1984). Variance reduction techniques for digital simulation. American Journal of Mathematical and Managemer,.~ Sciences ,1, 277-312.

INDEX TERMS

Primary Classification:
  G. Mathematics of Computing
  G.3 PROBABILITY AND STATISTICS
      Subjects: Multivariate statistics

Additional Classification:
  I. Computing Methodologies
  I.6 SIMULATION AND MODELING

General Terms:
Design, Experimentation, Measurement, Performance, Theory