Strategy-Proof Incentives for Predictions

TL;DR

This paper develops and analyzes three strategy-proof mechanisms for truthful prediction reporting that account for long-term strategic behavior, with one mechanism effectively ensuring truthfulness across most prediction scenarios.

Contribution

It introduces three novel mechanisms that are strategy-proof for non-myopic agents and provides a classification framework for prediction settings to identify truthful environments.

Findings

One mechanism achieves strategy-proofness in nearly all prediction settings.

Rules are formulated to distinguish truthful from untruthful prediction environments.

The proposed mechanism restores prompt truthfulness where existing mechanisms fail.

Abstract

Our aim is to design mechanisms that motivate all agents to reveal their predictions truthfully and promptly. For myopic agents, proper scoring rules induce truthfulness. However, as has been described in the literature, when agents take into account long-term effects of their actions, deception and reticence may appear. No simple rules exist to distinguish between the truthful and the untruthful situations, and a determination has been done in isolated cases only. This is of relevance to prediction markets, where the market value is a common prediction, and more generally in informal public prediction forums, such as stock-market estimates by analysts. We describe three mechanisms that are strategy-proof with non-myopic considerations, and show that one of them meets all our requirements from a mechanism in almost all prediction settings. We formulate rules to distinguish truthful from untruthful settings, and use them to extensively classify prediction settings with continuous outcomes. We show how our proposed mechanism restores prompt truthfulness where incumbent mechanisms fail, and offer guidelines to implementing it in a prediction market.