TLOB: Dual Attention Transformer Predicts Price Trends from Order Book Data
a year ago
- #Machine Learning
- #Market Efficiency
- #Quantitative Finance
- Introduces TLOB, a transformer-based model with dual attention for price trend prediction using Limit Order Book (LOB) data.
- Challenges the necessity of complex architectures by showing superior performance with a simple MLP-based approach.
- Proposes a new labeling method to remove horizon bias in predictions.
- Evaluates TLOB across four horizons using FI-2010 benchmark, NASDAQ, and Bitcoin datasets, outperforming state-of-the-art methods.
- Highlights declining stock price predictability over time, indicating growing market efficiency.
- Discusses the impact of transaction costs on trend classification and trading strategy profitability.
- Provides insights into the evolving landscape of stock price trend prediction and sets a foundation for future financial AI advancements.
- Code for TLOB is publicly released.