Data markets with dynamic arrival of buyers and sellers

We propose a market design solution for a market for distributed data. The main challenges addressed by our solution are (1) different data providers produce different databases that can be joined to produce answers for users' queries; (2) data providers have high fixed costs for producing their databases; and (3) buyers and sellers can arrive dynamically to the market. Our design relies on using a Markov chain with states corresponding to different numbers of allocated databases. The transition probabilities between different states are governed by the payments suggested by the market platform to the data providers. The main challenge in this setting is to guarantee dynamic incentive compatibility, i.e., to ensure that buyers and sellers are not incentivized to arrive late to the market or to misreport their costs or values. To achieve this, we disentangle the payments suggested by the market platform to the sellers from the posted prices exposed to the buyers. We prove that the buyer-optimal payments that are exposed to sellers are non-increasing which prevents late arrivals of sellers. Further, we demonstrate that the posted prices exposed to buyers constitute a martingale process (i.e., late arrivals lead to the same expected price). Finally, we show that our design guarantees zero expected average budget deficit and we perform a number of simulations to validate our model.