EarnHFT: Efficient Hierarchical Reinforcement Learning for High Frequency Trading

Authors: Molei Qin, Shuo Sun, Wentao Zhang, Haochong Xia , Xinrun Wang, Bo An

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments in various market trends on Crypto markets in a high-fidelity simulation trading environment, we demonstrate that Earn HFT significantly outperforms 6 state-of-art baselines in 3 popular financial criteria, exceeding the runner-up by 30% in profitability.
Researcher Affiliation Academia Molei Qin*, Shuo Sun*, Wentao Zhang, Haochong Xia, Xinrun Wang , Bo An Nanyang Technological University, Singapore {molei001,shuo003,haochong001}@e.ntu.edu.sg, {wentao.zhang,xinrun.wang,boan}@ntu.edu.sg
Pseudocode Yes Algorithm 1: Construction of Optimal Action Value; Algorithm 2: Efficient RL with Q-Teacher; Algorithm 3: Market Segmentation & Labelling
Open Source Code No The paper does not provide an explicit statement about releasing source code or a link to a code repository for their proposed method.
Open Datasets No The paper mentions using 'BTC/TUSD Sideways', 'BTC/USDT Sideways', 'ETH/USDT Bear', and 'GALA/USDT Bull' datasets. While it provides statistics in Table 1, it does not offer concrete access information (e.g., links, DOIs, or citations to publicly available datasets) for these specific datasets.
Dataset Splits Yes For dataset split, we use data from the last 9 days for testing, the penultimate 9 days for validation and the remaining for training on all 4 datasets.
Hardware Specification Yes We conduct all experiments on a 4090 GPU.
Software Dependencies No The paper mentions 'Adam is used as the optimizer for DDQN' and refers to the 'Cleanrl' library (Huang et al. 2022) for baselines. However, it does not provide specific version numbers for these or any other key software components used in their own implementation, which is required for reproducibility.
Experiment Setup Yes For the trading setting, the commission fee rate is 0 for BTCT and 0.02% for the remaining datasets following the policy of Binance. For the training setting, we choose β in Equaition 4 in list [ 90, 10, 30, 100] and run each β for 50 epochs, generating a total of 200 agents. Adam is used as the optimizer for DDQN.