Home Title

My project is to design and implement an agent for the SCM Prediction Challenge, a new challenge in 2007 in the Trading Agent Competition. This challenge is an offshoot from the TAC Supply Chain Management game (SCM).

In the SCM game, each agent is a computer manufacturer that must compete with other agents to buy components from suppliers, win orders from customers, and produce and deliver the ordered computers to customers. The agent that earns the most profit after a specified number of days wins. Every day, the agent receives from the customers requests for quotes (RFQs) for specified quantities of different types of computers (stock-keeping units or SKUs) to be delivered on specified dates, and must offer prices for those RFQs (without knowing the offer prices of other agents). The agent also receives offers from different component suppliers for the RFQs that the agent sent on the previous day, and must decide on which components to order and from which suppliers. The next day, for each RFQ, the customers will order from the agent that offers the lowest price for that RFQ.

A successful SCM agent must be able to predict both the prices that suppliers will offer for components and the price at which each customer RFQ today will be won tomorrow (or more generally, the probability of winning an offer for a given price). The prediction challenge was designed to improve strategies for these predictions (with the addition of predictions for future prices), and I have decided to participate in it. I hope to improve upon our SCM agent's current methods of prediction in my solution to the challenge, and if possible, transfer that solution into the SCM agent.