Title:Incentivizing Truthful Data Contributions in a Marketplace for Mean Estimation

Abstract:We study a data marketplace where a broker mediates transactions between buyers and contributors. Buyers estimate the mean mu of an unknown normal distribution N(mu, sigma^2), with valuations based on estimation error. Contributors, incurring different data collection costs, gather and report samples (not necessarily truthfully) to the broker, who sells subsets at varying prices and redistributes revenue. We model this as a mechanism design problem to maximize profit (revenue minus data costs) while ensuring individual rationality (participation benefits all), envy-freeness (no buyer prefers another allocation), budget balance (payments cover costs), and incentive compatibility (contributors truthfully report sufficient data).
We design a mechanism which satisfies these requirements. We first establish a connection between envy-free data pricing and ordered-item pricing (OIP) for unit-demand buyers and leverage OIP algorithms to determine the optimal data allocation and expected prices for buyers. The actual prices paid by buyers are centered around these expected prices, but also vary based on discrepancies in contributors' reported data. This variation is then passed on to the contributors via their payments. This scheme results in a Nash equilibrium (NE) where only the two lowest-cost contributors collect all the data and report it truthfully. Including this variation in the prices also helps us achieve individual rationality for buyers, as buyers will pay less if there are significant discrepancies in contributors' datasets.
To complement these findings, we prove a nearly matching upper bound on the maximum possible profit achievable in any NE of any mechanism, thus proving that our mechanism is essentially unimprovable. We also show that no nontrivial dominant-strategy incentive-compatible mechanism exists in this problem.