A Perspective and Proposal for the Initial Transient Problem in Simulation.
NORTH CAROLINA UNIV AT CHAPEL HILL
Pagination or Media Count:
This paper presents a perspective on the initial transient problem in steady-state simulation. In particular, it enumerates five generally accepted facts 1 Conditions prevailing at the beginning of a simulation influence sample paths. 2 The extent of influence is a function of the strength of autocorrelation. 3 Some initial conditions are less detrimental than others are. 4 Truncation reduces bias but usually increases variance. 5 So far no complete solution exists. The remainder of the paper describes a proposal for solving the problem. It relies on the relatively weak assumption that the conditional means in a stochastic process of interest are related linearly. An estimator of the steady-state mean is described which has considerably less bias than one can achieve via conventional truncation. An interval estimator is also described which follows from standard regression theory. A test for residual bias is presented which enables a user to judge whether or not sample data meet the minimal requirements for the proposed technique to apply. A second test allows a user to judge whether or not a more efficient estimation technique can be used. Author
- Statistics and Probability