Economic Regime Identification and Prediction in TAC SCM Using Sales and Procurement Information Frederik Hogenboom Erasmus Sch. of Econ. Erasmus University Rotterdam fhogenboom@ese.eur.nl Wolfgang Ketter Rotterdam Sch. of Mgmt. Erasmus University Rotterdam wketter@rsm.nl Jan van Dalen Rotterdam Sch. of Mgmt. Erasmus University Rotterdam jdalen@rsm.nl Uzay Kaymak Erasmus Sch. of Econ. Erasmus University Rotterdam kaymak@ese.eur.nl John Collins Computer Science and Engr. University of Minnesota jcollins@cs.umn.edu Alok Gupta Carlson Sch. of Mgmt. University of Minnesota agupta@csom.umn.edu Abstract Our research is focused on the effects of the addition of procurement information (offer prices) to a sales- based economic regime model. This model is used for strategic, tactical, and operational decision mak- ing in dynamic supply chains. We evaluate the perfor- mance of the regime model through experiments with the MinneTAC trading agent, which competes in the TAC SCM game. The new regime model has an over- all predictive performance which is equal to the per- formance of the existing model. Regime switches are predicted more accurately, whereas the prediction ac- curacy of dominant regimes does not improve. How- ever, because procurement information has been added to the model, the model has been enriched, which gives new opportunities for applications in the procurement market, such as procurement reserve pricing. Introduction Nowadays, markets are extremely competitive, and thus it is important to gain insight into the dynamics of supply chains and to research supply chain optimiza- tion possibilities, both for individual elements in the chain, as well as for the chain as a whole. For instance, in correctly predicting future market conditions, i.e., economic regimes (Ketter 2007), lie competitive advan- tages, e.g., one can anticipate on upcoming scarcities in the sales market by adjusting procurement policies and sales prices in advance. This can save for instance storage costs, and increase profits. Thus, one could benefit greatly from being able to make tactical and strategic decisions in an uncertain market, based on identified and predicted economic regimes. Combining techniques from computer science with economic theory to solve problems in economic environments contributes to novel approaches to existing problems. Ketter introduced an economic regime model, which is based on sales information (Ketter 2007). This model can be applied to any real or simulated market situa- tion. However, procurement information has not been Copyright c  2009, Association for the Advancement of Ar- tificial Intelligence (www.aaai.org). All rights reserved. used sofar for identifying and predicting regimes. This information is truly valuable for determining economic regimes, since it captures specific market characteris- tics. For instance, an increase in the amount of compo- nents sold in the procurement market could indicate an expected scarcity in the sales market, as manufacturers are building stocks. We present an extension to the regime model as in- troduced by Ketter, which is implemented in the Min- neTAC trading agent (Collins, Ketter, and Gini 2009). We investigate the effects of adding procurement infor- mation to this model. The performance of the regime model is evaluated through experiments on the quality of regime probability predictions, and checking correla- tions with existing market conditions. The MinneTAC trading agent has competed for sev- eral years in the Trading Agent Competition for Sup- ply Chain Management (TAC SCM). TAC SCM is an annual international competition for designing trad- ing agents for a simulated personal computer supply chain (Collins et al. 2005). The supply chain includes customers, traders, and suppliers and the computer market is divided into market segments. Traders have to procure components from suppliers and sell assem- bled computers to customers. Only limited information is available to traders, such as yesterday’s maximum and minimum market prices, which makes predicting the market a non-trivial task. TAC SCM has attracted researchers from all over the world, because its environment is designed in such a way that it contains many characteristics that can be found in real-life supply chains, such as the behavior of unpredictable opponents and interdependent chain entities. The simulated supply chain of the TAC SCM game offers many research opportunities into various subjects, such as price setting strategies and prediction strategies for competitor behavior or market character- istics and developments. The paper is organized as follows. First, we con- tinue with introducing the existing economic sales- based regime model. Subsequently, we define and eval- uate a new economic regime framework based on both