When AI Advisors Agree: Distinguishing Forecast Risk from Forecast Reliability

Key points

• Agreement can indicate useful shared evidence, but it can also reflect an asset that was easier to forecast in the first place.1
• We examine this distinction using 928 completed first-quarter forecasts for 232 assets across four production batches of iPulse AI.1
• The relationship between disagreement and additional forecast error is not resolved after accounting for visible asset risk.1
• These results distinguish three questions often collapsed into one confidence score: how risky the asset is, whether advisors agree on its direction, and how precisely they forecast its return.1

Sources (1)

• [1]When AI Advisors Agree: Distinguishing Forecast Risk from Forecast Reliability
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 11:34 AM

  If seven AI advisors give similar forecasts, should we trust them more?

  Agreement can indicate useful shared evidence, but it can also reflect an asset that was easier to forecast in the first place.

Extractive summary: sentences quoted from the sources.