LOT: Combining LOB and OHCLV Data for Improved Stock Trading

This work introduces LOT, a novel Deep Reinforcement Learning (DRL) architecture that combines Limited Order Book (LOB) and open, high, low, close, volume (OHCLV) data for highly efficient and stable automatic cryptocurrency trading. The LOT model was developed to select optimal trading strategies and maximize profits by leveraging LOB and OHCLV data and additional market metrics. The architecture utilizes convolutional and recurrent layers to capture and retain patterns from these diverse data sources, providing a more comprehensive understanding of market dynamics. During the training phase, the model achieved a profit of 0.15% over an 83-minute trading session, demonstrating its early effectiveness. During the testing stage, the model achieved an average profit of 0.7% over a 7 -hour trading period, further highlighting its potential. As of the final result after training, the LOT model delivered an impressive average daily profit of 2.56%, surpassing the next best model by 13 % and outperforming the average of other models by 83%. Additionally, compared to traditional buy-and-hold strategies, which generated profits averaging 0.001%, the LOT model achieved a profit of 0.15% over the same trading window, demonstrating a substantial improvement.