Abstract: Based on the auction formalism in Onsale, Inc., an Internet-based auction house, we study the following optimization problem: here a given total number of iterms should be sold in a given time horizon, which is divided into serveral time periods and at each period there is an auction to sold the items offered optimally. The remaining items at each period can be sold at the next period until the time horizon is ended. It is formulated as a Markov decision process and its optimality equuation is solved numerically. It is shown that the expected total profit is superior to the original one.

Key words: optimal mechanism, Internet auction, Markov decision process