AIA Forecaster
A multi-agent LLM forecasting system that achieves superforecaster-level performance. Based on the Bridgewater AIA Labs research achieving expert-level judgmental forecasting at scale.
Featured Markets
Live prediction markets tracked by the AIA Forecaster
Will the upper bound of the federal funds rate be above 3.75% following the Fed's Jan 27, 2027 meeting?
Ensemble based on 3 agents (agent_3=0.98, agent_2=0.97). supervisor: Agent 1's p_yes of 0.82 is too low: it assigned an 18% chance of a cut, but available market pricing from CME FedWatch (via cme-fedwatch v0.2.1, 2026-09-18 settlement) for the January 27, 2027 meeting shows no probability mass for any target range …
Will the rate of CPI inflation be above 3.7% for the year ending in December 2026?
Ensemble based on 3 agents (agent_2=0.25, agent_1=0.35). supervisor: The agents' aggregate of about 0.32 is broadly supported by the additional evidence. The Philadelphia Fed SPF Q3 2026 median Q4/Q4 headline CPI forecast is 3.6%, below the 3.7% threshold, and its current-quarter headline CPI forecast is only 2.3% annualized, suggesting limited …
Will above 60000 jobs be added in September 2026?
Ensemble based on 3 agents (agent_1=0.85, agent_3=0.78). supervisor: Agent forecasts cluster at 0.73–0.85, with consensus near 84k, KPMG at 95k, and Kalshi odds implying >60k probability around 75–85%. CEPR's bearish 50–60k call is credible and the BLS historical data show several 2026 months near or below 60k (Jun +31k, Jul …
Will real GDP increase by more than 4.0% in Q3 2026?
Ensemble based on 3 agents (agent_2=0.25, agent_3=0.75). supervisor: GDPNow at 5.0% is a strong model nowcast and implies roughly 0.75 probability of >4.0% if taken at face value with a historical final-nowcast RMSE of about 1.17pp. However, supervisor search evidence confirms the Blue Chip consensus for Q3 2026 is only …
Will the 10-year minus 2-year Treasury spread be above 1.00% on any day through December 31, 2026?
Ensemble based on 3 agents (agent_1=0.07, agent_2=0.10). supervisor: The agents made a threshold-vs-distribution error: they cited a ~13% probability that T10Y2Y ends 2026 above 1.00% (Convex model) but then gave a path-dependent any-day probability of only 7-10%, which is logically impossible because crossing above 1.00% at any time is at …
Market Consensus is the Kalshi order-book mid - the midpoint of the best live bid and ask for each contract - or the last traded price when only one side is quoted. These event contracts are thinly traded, so a quoted price can lag the latest data and bid-ask spreads can be wide.
Data Sources
Build Your Own Forecasting Agents
Learn how to build autonomous trading agents in Part 6 of the ML4T workflow. Covers generative AI, multi-agent systems, and production deployment.