Research on Bidding Strategies in Online Multi-Item Auctions Based on Markov Decision Process

This paper explores the procurement strategy of a company that, in order to meet market demand, builds its inventory by participating in multiple online auctions for homogeneous items. To characterize the early and late bidding behaviors of auction participants, a time variable is introduced, and a non-homogeneous Poisson process is established to model bidder arrivals. The winning probabilities are derived for auctions with either a single item or multiple items available for purchase, and a Markov decision process model is developed to determine the optimal bidding strategy that maximizes the company’s profit. Under the assumption of identical conditions and unit demand, it is demonstrated that the winning probability in auctions with a single item is lower than in auctions with multiple identical items, where only one item can be purchased per auction. Numerical analysis reveals that the cumulative intensity function has a significant impact on the winning probability, and errors in estimating its parameters may lead to the failure of the bidding strategy.