Researchers have developed UQ-LOB, a novel uncertainty quantification module designed to enhance deep learning models for limit order book (LOB) mid-price forecasting. This module attaches to existing LOB encoders and provides calibrated predictions with a scalar confidence score, allowing for selective prediction based on forecast reliability. Tested on billions of LOB events across seven cryptocurrency assets, UQ-LOB demonstrated improved directional accuracy, particularly when focusing on the most confident predictions, achieving a directional F1 score of up to 0.88 at a 5-second horizon. AI
IMPACT Enhances reliability of AI-driven trading predictions by quantifying forecast confidence.
RANK_REASON Research paper detailing a new method for uncertainty quantification in AI forecasting. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Derrick Gilchrist
- Edward Manoharan
- Gotit.pub
- Hugging Face
- ScienceCast
- UQ-LOB
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
