Research map / Label-efficient and robust learning / Noisy labels
The Label Horizon Paradox: Rethinking Supervision Targets in Financial Forecasting
Chen-Hui Song, Shuoling Liu, Liyuan Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train on learned intermediate-horizon labels instead of the exact target horizon | 10 backbones · CSI 300/500/1000 daily | IC: 0.637 -> 0.720 (CSI 300 LSTM) | Improved |
| train on minute-level horizons earlier than the next-day target (ablation) | LSTM · CSI 500 Scenario 1 (daily) | out-of-sample IC | Improved |
| train on target horizon itself in 30-minute window (ablation) | LSTM · CSI 500 Scenario 2 (30-min) | out-of-sample IC | Improved |
| use intermediate horizon label in 90-minute window (ablation) | LSTM · CSI 500 Scenario 3 (90-min) | out-of-sample IC | Improved |
| replace learned horizon weights with naive averaging of labels (ablation) | Bi-level framework · CSI 500 Scenario 1 | IC (x10): 1.029 -> 0.969 | Hurt |
| replace learned horizon weights with equal-weight multi-task learning (ablation) | Bi-level framework · CSI 500 Scenario 1 | IC (x10): 1.029 -> 0.932 | Hurt |
| replace learned horizon weights with naive averaging in 90-minute scenario (ablation) | Bi-level framework · CSI 500 Scenario 3 | ICIR: 1.095 -> 1.125 | Mixed |
| add warm-up phase before bi-level optimization (ablation) | Bi-level framework · CSI 500 LSTM Scenario 3 | learned horizon weights | Improved |
| use bi-level method in 30-minute window | LSTM · CSI 500 Scenario 2 | IC (x10) | No clear effect |
Closest papers by meaning
- Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context (AAAI 2024 · similarity 0.85)
- Towards Financial World Modeling (NeurIPS 2026 · similarity 0.85)
- Retrieval-Augmented Diffusion Modeling for Stochastic Discount Factor Portfolios (NeurIPS 2026 · similarity 0.83)
- Abstain Mask Retain Core: Time Series Prediction by Adaptive Masking Loss with Representation Consistency (NeurIPS 2025 · similarity 0.83)
- Generating Financial Time Series by Matching Random Convolutional Features (NeurIPS 2026 · similarity 0.83)
- Filtered not Mixed: Filtering-Based Online Gating for Mixture of Large Language Models (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
