This report presents a novel approach to cryptocurrency trading using a Transformer-based Deep Reinforcement Learning (DRL) agent, yielding improved adaptability and profitability.
By combining Transformers, Double DQN, Noisy Networks, and LoRA for Test-Time Training, the agent showcases significant enhancements in its trading strategy and profitability on historical BTC/USDT data.
Preliminary results indicate that this innovative method has the potential to extend to more complex datasets and multiple asset classes, which can be beneficial for institutional-grade trading strategies.
Transformers are adept at analyzing time-series data in financial markets, enabling traders to anticipate significant price moves before they occur, outperforming traditional quant methods.
Collection
[
|
...
]