Adaptivespectralmethodsforsimulationoutputanalysis.pdfVIP

Philip Heidelberger Peter D. Welch Adaptive Spectral Methods for Simulation Output Analysis Thispaper addresses two central problems in simulation methodology: the generation of conjidence intervalsfor the steady state meansof the output sequences and the sequential ofusethese conjidenceintervals to control the run length. The variance of the sample meanof a covariance stationary process is given approximatelyby p(O)lN, where p(f) is the f and N is the sample size. Inearlieran paper we developed a method ofconfidence interval spectral density at frequency generation basedontheestimation of p(0)throughtheleastsquaresjit of a quadratictothelogarithm of the periodogram. This method was applied in a run length Controlprocedure to a sequence of batched means. As the run length increased the batch means were rebatched into larger batch sizes so as to limit storage requirements. In this rebatching the shapeof the spectral density changes,gradually becoming flat asN increases. Quadraticswere chosen as a compromise between small sample bias and large sample stability. In this paper we consider smoothing techniques which adapt to the changing spectral shape in an attempt to improve both the small and large sample behavior of the method. The techniques consideredare polynomial smoothing with the degree selected sequentially using standard regression stat