Repository files navigation

DeepFolio is anOpenAI Sora-inspired Diffusion-Transformer (DiT) framework forjoint portfolio optimization and best execution, designed tomaximize Sharpe ratio without explicit return forecasts. It leverages:

• Transformer to capture asset dependencies and encode market conditions.
• Diffusion Models to filter market noise and generate bothrobust allocation weights andoptimized trading trajectories.
• End-to-End Strategy Execution to reduce information loss betweenstrategy design and execution implementation, ensuring optimal real-world performance.

✅Unified Portfolio & Execution Optimization – Bridges the gap between portfolio construction and trade execution.
✅Diffusion-Based Portfolio Generation – Generatesadaptive, robust asset allocations without relying on explicit return forecasts.
✅Market-Aware Execution Path Modeling – UsesDiffusion Models to optimizeexecution trajectories, reducing slippage and market impact.
✅Scenario-Based Adaptation – Dynamically adjusts strategies forhigh/low volatility regimes, liquidity shifts, and market anomalies.
✅Transaction Cost-Aware Optimization – IntegratesTCA (Transaction Cost Analysis) into optimization, minimizing execution costs.

DeepFolio consists oftwo core modules:

• Transformer Encoder extracts asset relationships, learning market structure.
• Diffusion Model generates optimal portfolio weights, ensuring robustness under different conditions.

• Transformer encodes market microstructure (LOB, liquidity, volatility).
• Diffusion Model optimizes execution paths to minimize market impact and transaction costs.

📌Pipeline Overview:

For detailed documentation, please visit ourdocumentation site.

We welcome contributions! Please see ourcontributing guidelines for more details.

This project is licensed under the BSD-2-Clause License- see theLICENSE file for details.

[1]Damian Kisiel, Denise Gorse (2022).Portfolio Transformer for Attention-Based Asset AllocationarXiv:2206.03246 [q-fin.PM]

• This package leverages the power of TensorFlow for efficient portfolio optimization.
• Thanks to the financial machine learning community for inspiring many of the implemented methods.