Forecast Collapse in Time-Series Foundation Models

2026-08-14

AI Summary

This study addresses the issues of prediction collapse and ranking failure in temporal foundation models when applied to low-predictability targets, revealing that these limitations stem from an inherent trade-off between calibration and ranking capability. To overcome this, we propose CalibRank, a joint optimization objective that effectively balances both aspects while addressing blind spots in conventional evaluation metrics. Experiments on the Finance1K dataset demonstrate that our method nearly triples cross-sectional correlation while maintaining stable prediction magnitudes. These results indicate significantly enhanced ranking performance and robustness in financial time series forecasting. By reconciling the calibration-ranking dilemma, this work establishes a novel paradigm for overcoming critical bottlenecks in deploying foundation models for real-world financial applications.

Abstract

When forecasting hourly returns for 1,000 US equities, we observe an unexpected phenomenon: predictions become nearly flat and show poor stock ranking, as measured by cross-sectional correlation. We call this forecast collapse. Surprisingly, the phenomenon largely disappears when forecasting trading volume under the same setting. We investigate forecast collapse across time-series foundation models (TSFMs), twelve deep-learning forecasting models, and 97 public benchmark configurations, and find that it is closely tied to target predictability. We identify two distinct reasons behind it: low predictability limits the amplitude of calibrated point forecasts, while per-series objectives leave cross-series structure unidentified. These findings reveal a calibration-ranking tradeoff: optimizing squared error leads to flat predictions, whereas directly optimizing cross-sectional correlation improves ranking but can inflate forecast amplitude by more than an order of magnitude. To address this tradeoff, we introduce CalibRank, a simple objective that balances calibration and ranking. On Finance1K, CalibRank nearly triples cross-sectional correlation while keeping amplitude close to the target, and improves correlation on all tested models. Our results reveal a blind spot in conventional time-series evaluation: per-series metrics can hide failures in cross-series structure needed by downstream decisions.

Problem

• Forecast Collapse
• Time-Series Foundation Models
• Cross-sectional Correlation
• Calibration-Ranking Tradeoff
• Target Predictability

Innovation

• Forecast Collapse
• CalibRank
• Calibration-Ranking Tradeoff
• Time-Series Foundation Models
• Cross-Sectional Correlation