A Sequential Stopping Rule for a Steady-State Simulation Based on Time-Series Forecasting

Abstract

A sequential stopping procedure should collect enough steady-state data to overwhelm the influence of initial transient bias without requiring initial data truncation. The initial transient negatively affects the efficiency of the sequential procedure, but from a practical point of view, eliminating the difficulty of determining the data truncation point can lead to a more easily implemented algorithm for determining the appropriate length of a simulation run. A sequential stopping rule is presented that uses a time-series forecasting procedure to determine appropriate trade-offs between the efficiency and simplicity of the estimate of cycle time for a relevant constant mean process. Results show that the proposed sequential stopping rule terminates a simulation output process at a point when a stable estimate is obtained. Furthermore, the rule performs as well as the crossings-of-means data truncation technique yet is easier to implement.