Identifying and Predicting Economic Regimes in TAC SCM

In this thesis, the effects of adding procurement information to a sales-based regime model, which is used for predicting price trends in a simulated supply chain, are researched. This supply chain is simulated in the TAC SCM game, which is an annual international competition held for several years, where researchers from around the world submit their artificial trading agents. The regime model extended in this thesis is used by the MinneTAC agent of the University of Minnesota. We find that component offer prices can be used to extend the regime model, which is currently based on a one-dimensional Gaussian Mixture Model where probabilities are clustered. The resulting clusters hold as regimes. Extending the model with a new dimension results in newly defined regime clusters. Implementing the new regime model, MinneTAC increases its customer orders significantly. However, because the agent configuration shows a structural error in predicting future price trends – possibly due to an insufficient pricing mechanism – we have strong indications that our new approach leads to lower profits, although the decrease of the amount of cash at the end of a game is not significant. We believe that this decrease of profits can be tackled in the future by research into price trend prediction in the newly defined regime model. i