Reproduce “History Is Not Enough: An Adaptive Dataflow System for Financial Time-Series Synthesis” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Haochong Xia, Yao Long Teng, Regan Tan, Molei Qin, Xinrun Wang, Bo An

Researchers from Nanyang Technological University and Singapore Management University developed a drift-aware adaptive dataflow system for financial time-series synthesis, integrating data manipulation, adaptive scheduling, and learning-based control. This system consistently improved forecasting accuracy and enhanced risk-adjusted returns in reinforcement learning trading tasks by effectively addressing concept drift.

Key insights

  • Maintaining financial fidelity and specific stylized facts in augmented data is critical, necessitating operations that ensure K-line consistency and reflect market properties.
  • An adaptive, learning-guided planner and scheduler for data augmentation, responsive to model performance and overfitting, is essential for robustly countering concept drift.
  • Multi-stock mix-up operations, guided by cointegration, provide valuable cross-asset information that aids in learning more generalizable temporal dynamics.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.