Beyond Forecasting: The Belief-to-Trade Layer in Prediction-Market Agents

This paper argues that prediction-market agents require more than accurate and calibrated forecasts: they must convert beliefs into trading decisions, including position selection and risk management. The authors introduce Raven-Agent, described as an autonomous trading agent for prediction markets. In a controlled replay over an archived decision set, Raven-Agent reportedly achieved the only positive return and the only positive risk-adjusted return among the tested policies. The implementation has been released as predict-raven on GitHub. The abstract does not provide the return values, benchmark composition, or detailed experimental protocol.