My Project


Ad Auction

My project is to build a trading agent for the trading agent competition in ad auction (TAC/AA). TAC/AA provides a simulated environment for ad auction, or keyword auction. Here is a link that briefly describes Google's keyword auction. It brings billions of dollars to Google ($21 billion in 2008, see this wikipedia article obout Google AdWords.), but the dynamics of these auctions are not well understood yet.

Overview of TAC/AA

In TAC/AA, there are 60 simulated days with 8 agents competing for 16 keywords combination, or queries. The 8 agents attempts to sell the same set of products by bidding on the 16 queries but may have different specialty or ad qualities. Each day the agent will need to submit a bid bundle, which specifies the bid price and spending limit for each query. They can also specify an overall budget for the entire ad compaign. At the end of each day, the agents will receive a sales report and a query report, which contains some information about the ad auction such as, average position, average CPC, impressions, number of clicks, conversions(purchases), etc. of each query. In order to enforce some marginal return on bids, agents have different capacities (i.e. the maximum product they can sell over a 5 day window), and will be penalized for overselling. For more details on this topic, please refer to the spec of the game.