Department of Mathematics

Dow Chemical

Proposed Project for the MSU Industrial Math Students

Phase I: Replicating a Commodity Market Simulation Using Agent Based Modeling Techniques

The ability to reasonably forecast the prices of the chemical commodity market is of great importance to many departments within The Dow Chemical Company. This project with be the first phase of two in the development of such a pricing model. The focus of this project will be to mimic the existing iThink simulation solution from John D. Sterman's book "Business Dynamics: Systems Thinking and Modeling for a Complex world" using the REPAST toolkit developed largely by Argonne National Laboratory. REPAST is an Agent Based Modeling development toolkit that is free for download from Sourceforge.net and will provide the platform on which to build the simulations.

The deliverables of this project include: a literature review on the ability for dynamic simulation models (systems dynamics or agent based modeling) to represent real world commodity supply/demand cycles..both the large cycle (supply/demand) and small interim cycles (stocking and destocking phenomena), a comparison of historical data with "historical" simulations, and the incorporation of good statistical forecasts (supply, demand, raw material costs, etc.), potentially with feedback, into the simulations.

Phase II: Modeling the Supply/Demand Dynamics in a Commodity Chemical Market

This project will continue the work of Phase I: Replicating a commodity market simulation using Agent Based Modeling techniques, by taking the model developed and modifying it to represent the specific supply/demand markets of chemicals used by The Dow Chemical Company.

The final deliverable would be an Agent Based Model created using REPAST software that represents real Dow data for the supply/demand dynamics in a commodity chemical market. Various structural phenomena and an iThink model will already have been identified by an internal Dow team that will be used to build the REPAST model using Agent Based Modeling.