Research map / Architectures and efficiency / Attention and transformers
Signature-Informed Transformer for Asset Allocation
Yoontae Hwang, Stefan Zohren
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 |
|---|---|---|---|
| add CVaR objective instead of risk-neutral mean return (ablation) | SIT · S&P100 40 assets | Sharpe ratio: 0.5691 -> 0.6717 | Improved |
| add CVaR objective (ablation) | SIT · S&P100 50 assets | Sharpe ratio: 0.5923 -> 0.7715 | Improved |
| add asset-wise signature-informed attention (ablation) | SIT · S&P100 40 assets | Sharpe ratio: 0.5284 -> 0.6717 | Improved |
| add signature-derived financial bias to attention (ablation) | SIT · S&P100 40 assets | Sharpe ratio: 0.6045 -> 0.6717 | Improved |
| add learnable gate gamma (ablation) | SIT · S&P100 40 assets | Sharpe ratio: 0.5251 -> 0.6717 | Improved |
| add learnable gate gamma (ablation) | SIT · S&P100 50 assets | Sharpe ratio: 0.5831 -> 0.7715 | Improved |
| add financial bias (ablation) | SIT · S&P100 50 assets | Maximum drawdown: 0.3224 -> 0.3271 | Hurt |
| add asset attention (ablation) | SIT · S&P100 40 assets | Maximum drawdown: 0.3342 -> 0.3611 | Hurt |
| use decision-focused learning instead of predict-then-optimize | eight forecasting backbones · S&P100 40 and 50 assets | Sharpe ratio | Improved |
| increase softmax temperature tau from low to about 1.3 | SIT · S&P100 40 and 50 assets | Sharpe ratio | Mixed |
| raise transaction cost from 0 to 10 bps | SIT · S&P100 40 and 50 assets | Sharpe ratio: about 0.03-0.04 | Hurt |
Closest papers by meaning
- Generating Financial Time Series by Matching Random Convolutional Features (NeurIPS 2026 · similarity 0.82)
- Retrieval-Augmented Diffusion Modeling for Stochastic Discount Factor Portfolios (NeurIPS 2026 · similarity 0.82)
- Probabilistic Signature Inversion: Learning Conditional Distributions from Truncated Signatures (NeurIPS 2026 · similarity 0.82)
- DHMoE: Diffusion Generated Hierarchical Multi-Granular Expertise for Stock Prediction (AAAI 2025 · similarity 0.81)
- Time Travel is Cheating: Going Live with DeepFund for Real-Time Fund Investment Benchmarking (NeurIPS 2025 · similarity 0.81)
- Towards Financial World Modeling (NeurIPS 2026 · similarity 0.81)
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.
