A recent benchmark study found that prominent time-series foundation models, including TimesFM 3.0, Chronos-2, and Toto 2.0, failed to outperform simple baseline strategies when applied to market time series data. The experiments, which involved zero-shot evaluations on US stocks and ETFs across daily, hourly, and 5-minute intervals, showed that these advanced models achieved directional accuracy around chance levels. In some cases, basic strategies like predicting the market would always go up, or even simulated random walks, performed comparably or better than the sophisticated models. AI
IMPACT Suggests current foundation models for time-series forecasting may not be suitable for complex financial market prediction.
RANK_REASON Research paper evaluating AI models against baselines. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
