Generative model of bid sequences in lowest unique bid auctions

Lowest unique bid auction (LUBA) sites are gaining popularity on the Internet in recent years. In this paper, we study LUBA with resubmission in discrete bid spaces. A long-standing goal in the field of Internet auction is to develop agents that can perceive and understand the strategy information behind the mechanism and can guide us to behave in a fast, frugal and smart way. We marry ideas from recurrent neural network and data to learn a generative model for generating winning bid sequences. A sequence of winning bids in Internet auctions can be viewed as a sequence of events and modeled by generative models. We learn a model that is able to capture the long dependencies in a winning bid sequence. The generated data obtained from our model and the ground truth dataset share similar distributions.