A Note on Verification of Computer Simulation Models

Abstract

This paper establishes an argument that questions the validity of one “test” of goodness-of-fit for the simulated time path of a single endogenous variable in a simultaneous, perhaps dynamic, econometric model. The test was suggested by Cyert and Cohen, and consists of two parts, within the context of a regression of the actual series on the generated series: a test that the intercept of this regression differs significantly from zero and a test that the slope of this regression differs significantly from one. Presumably, the intuition underlying the test is that if the simulation model is a good one this regression should be a 45° line through the origin. The paper's primary purpose is to demonstrate that this intuition is wrong in general for the case of “stochastic simulation.”