Peer Prediction

A mechanism that incentivizes truthful reporting by rewarding participants based on how well their predictions align with others’ beliefs.

What is Peer Prediction?

Peer prediction is an information aggregation mechanism that incentivizes participants to report their true beliefs by rewarding them based on how accurately their predictions match the distribution of others’ beliefs, rather than the actual outcome of an event. As Scott Kominers describes in the transcript, participants are asked for their own forecast (e.g., will a candidate win?) and their estimate of what others believe, using the latter to cross-check truthfulness. This method is particularly useful in thin markets or for subjective questions where outcomes are hard to verify, unlike prediction markets that require a definitive resolution.

For example, the transcript references a study by Hussam, Rigol, and Roth, where peer prediction was used in a developing country to identify successful micro-entrepreneurs by asking community members who they and others thought would succeed. Unlike prediction markets, peer prediction allows immediate payouts based on belief alignment, not waiting for event resolution, making it practical for contexts with delayed or unclear outcomes. It leverages Bayesian reasoning to ensure truth-telling, as participants’ rewards depend on accurately gauging the crowd’s beliefs.

Peer prediction complements prediction markets by addressing scenarios where thick markets or objective outcomes are absent, such as small-scale or subjective forecasts. Its integration with blockchain could enhance trust through transparent reward distribution, but it’s less suited for high-stakes, objective events like elections, where prediction markets excel due to their financial incentives and clear resolutions.