Do Your Own Research: Learning to Forecast by Learning to Search

2026-10-01

AI Summary

This study addresses the limitation of existing forecasting models in leveraging reward signals to optimize dynamic information acquisition capabilities. To bridge this gap, we construct an agent-based forecasting environment grounded in Polymarket, marking the first integration of dynamic information retrieval into the training loop. Using Qwen3.5 as the backbone, we employ the GRPO reinforcement learning algorithm coupled with a Brier score-based reward mechanism and hierarchical leakage filtering techniques, enabling the model to autonomously search for and evaluate real-time evidence during training. By directly aligning search proficiency with prediction-driven rewards, our approach improves calibration by 30–40% and reduces inference costs by 95%, while surpassing state-of-the-art proprietary models on challenging forecasting tasks.

Abstract

Outcome-based reinforcement learning can train language models to forecast real-world events, but prior forecasting work either freezes research context before training or deploys agentic research only at test time, so the skill of gathering evidence is never shaped by the reward. We introduce an agentic forecasting environment, dataset, and harness built from 2,100+ resolved Polymarket questions; the agent acquires its own context at rollout time (web search, page reading, and financial time series, all restricted by layered leak filtering to information published before each question's cutoff), and we train Qwen3.5-35B-A3B (3B active parameters) on it with single-epoch GRPO under a Brier-score reward. Training changes how the agent interacts with information: calibration improves 30-40%, and search attempts fall from 3.8 to 2.25 per rollout as evidence discipline is learned. Evaluated in an identical harness against four frontier models, the trained policy also finishes ahead of every frontier model tested at evidence-based forecasting, including Claude Opus 4.5 (soft-Brier 0.254 vs. 0.256, n=265), at about 5% of the inference cost, and its margin is widest on the hardest questions, the ones the crowd itself had not decided. We release the environment, dataset, and per-rollout records as a reusable harness for temporal forecasting agents.

Problem

• forecasting
• reinforcement learning
• evidence gathering
• language models
• agentic search

Innovation

• Agentic Forecasting
• Reinforcement Learning
• GRPO
• Information Retrieval
• Leak Filtering