エージェント・レベルでの妥当性評価 : 株式市場シミュレーションによるケース・スタディ

概要

論文の詳細を見る
ABM (Agent Based Model/Modeling) is a simulation method, in which one makes a model as a collection of autonomous agents and investigate spontaneous behavior of such a model. It has high flexibility in description, so ABM simulations sometimes show more realistic behavior than EBM (Equation Based Model/Modeling). However ABM has such a disadvantage that a model builder easily commits arbitrary modelin g and cannot find reliable explanation of causal relation between parameter settings and simulation outcomes thereof. In order to resolve this disadvantage, we consider it critical to validate ABM simulation at agent level. We performed Monte Carlo simulations using two stock market models which are implemented through ABM. Although the behaviors of these two models look similar at system level, yet we can make the differences obvious at agent level by comparing action-entropy. We conclude this method is a useful criterion to estimate model quality and can contribute to ABM method by providing a validation-tool at agent level.
日本シミュレーション学会の論文
2000-03-15