New proposals for the validation of trace-driven imulations

Abstract

Simulations try to mimick reality. Hence, it is desirable that simulated values resemble the values observed in real life. Although the problem how to validate a simulation model has received quite a lot of attention, even recent literature offers no generally accepted standard method. This paper proposes a general approach. For trace-driven simulation, three different objectives are formulated. At the one extreme, the individual outcomes of a real life process and the simulated process should agree. An intermediate objective is that the outcomes of the two processes should have a symmetric joint distribution. And at the other extreme, only the marginal distributions of both outcomes have to be comparable. For all three objectives, validation measures and/or tests are proposed. For the last objective, a simulation study is presented, comparing two competiting tests