The CPC Seminar Series takes place between October and June, all seminars are free to attend and no registration is required. If you would like to present please contact firstname.lastname@example.org.
4th April 2014 3pm, University of Southampton, 58/1007
Bogumil Kaminski, Division of Decision Analysis and Support, Institute of Econometrics, Warsaw School of Economics
In the presentation we propose a method of hypothesis verification for stochastic simulation models using samples of their outputs. A Bayesian procedure is developed to evaluate the probability that the hypothesis under investigation is true for a single simulation input. Next, we prove that the only coherent way of aggregating such probabilities over different simulation inputs is by averaging. Moreover, conditions for averaging to give consistent and asymptotically unbiased evaluation of input space proportion, for which the tested hypothesis is true, are given. Finally, it is shown that in hypothesis verification, under limited simulation effort capacity assumption, there is a bias-variance trade-off between the number of input points sampled and sample size per point. The obtained theoretical results are illustrated by analysis of a simulation model presented in paper "Volatility Clustering in Financial Markets: Empirical Facts and Agent-Based Models" by Cont (2007).