Information Elicitation Mechanisms for Statistical Estimation.

摘要: We study learning statistical properties from strategic agents with private information. In this problem, must be incentivized to truthfully reveal their information even when it cannot directly verified. Moreover, the reported by aggregated into a estimate. two fundamental properties: estimating mean of an unknown Gaussian, and linear regression Gaussian error. The each agent is one point in Euclidean space.Our main results are mechanisms for these problems which optimally aggregate truth-telling equilibrium:• A minimal (non-revelation) mechanism large populations — only need report value, but that value not point.• small non-minimal answer more than question.These “informed truthful” where reporting unaltered data (truth-telling) 1) forms strict Bayesian Nash equilibrium 2) has strictly higher welfare any oblivious agents' strategies independent signals. also show revelation (each reports her signal) restricted setting use impossibility result prove necessity restriction.We build upon peer prediction literature single-question setting; however, most previous work area focuses on discrete signals, whereas our inherently continuous, we further simplify reports.