Awesome

The HIST framework for stock trend forecasting

The implementation of the paper "HIST: A Graph-based Framework for Stock Trend Forecasting via Mining Concept-Oriented Shared Information".

Environment

1. Install python3.7, 3.8 or 3.9.
2. Install the requirements in requirements.txt.
3. Install the quantitative investment platform Qlib and download the data from Qlib:

   # install Qlib from source
   pip install --upgrade  cython
   git clone https://github.com/microsoft/qlib.git && cd qlib
   python setup.py install
   
   # Download the stock features of Alpha360 from Qlib
   python scripts/get_data.py qlib_data --target_dir ~/.qlib/qlib_data/cn_data --region cn --version v2
   

Reproduce the stock trend forecasting results

git clone https://github.com/Wentao-Xu/HIST.git
cd HIST
mkdir output

Reproduce our HIST framework

# CSI 100
python learn.py --model_name HIST --data_set csi100 --hidden_size 128 --num_layers 2 --outdir ./output/csi100_HIST

# CSI 300
python learn.py --model_name HIST --data_set csi300 --hidden_size 128 --num_layers 2 --outdir ./output/csi300_HIST

Reproduce the baselines

• MLP

# MLP on CSI 100
python learn.py --model_name MLP --data_set csi100 --hidden_size 512 --num_layers 3 --outdir ./output/csi100_MLP

# MLP on CSI 300
python learn.py --model_name MLP --data_set csi300 --hidden_size 512 --num_layers 3 --outdir ./output/csi300_MLP

• LSTM

# LSTM on CSI 100
python learn.py --model_name LSTM --data_set csi100 --hidden_size 128 --num_layers 2 --outdir ./output/csi100_LSTM

# LSTM on CSI 300
python learn.py --model_name LSTM --data_set csi300 --hidden_size 128 --num_layers 2 --outdir ./output/csi300_LSTM

• GRU

# GRU on CSI 100
python learn.py --model_name GRU --data_set csi100 --hidden_size 128 --num_layers 2 --outdir ./output/csi100_GRU

# GRU on CSI 300
python learn.py --model_name GRU --data_set csi300 --hidden_size 64 --num_layers 2 --outdir ./output/csi300_GRU

• SFM

# SFM on CSI 100
python learn.py --model_name SFM --data_set csi100 --hidden_size 64 --num_layers 2 --outdir ./output/csi100_SFM

# SFM on CSI 300
python learn.py --model_name SFM --data_set csi300 --hidden_size 128 --num_layers 2 --outdir ./output/csi300_SFM

• GATs

# GATs on CSI 100
python learn.py --model_name GATs --data_set csi100 --hidden_size 128 --num_layers 2 --outdir ./output/csi100_GATs

# GATs on CSI 300
python learn.py --model_name GATs --data_set csi300 --hidden_size 64 --num_layers 2 --outdir ./output/csi300_GATs

• ALSTM

# ALSTM on CSI 100
python learn.py --model_name ALSTM --data_set csi100 --hidden_size 64 --num_layers 2 --outdir ./output/csi100_ALSTM

# ALSTM on CSI 300
python learn.py --model_name ALSTM --data_set csi300 --hidden_size 128 --num_layers 2 --outdir ./output/csi300_ALSTM

• Transformer

# Transformer on CSI 100
python learn.py --model_name Transformer --data_set csi100 --hidden_size 32 --num_layers 3 --outdir ./output/csi100_Transformer

# Transformer on CSI 300
python learn.py --model_name Transformer --data_set csi300 --hidden_size 32 --num_layers 3 --outdir ./output/csi300_Transformer

• ALSTM+TRA

  We reproduce the ALSTM+TRA with its source code.

Citation

Please cite the following paper if you use this code in your work.

@article{xu2021hist,
  title={HIST: A Graph-based Framework for Stock Trend Forecasting via Mining Concept-Oriented Shared Information},
  author={Xu, Wentao and Liu, Weiqing and Wang, Lewen and Xia, Yingce and Bian, Jiang and Yin, Jian and Liu, Tie-Yan},
  journal={arXiv preprint arXiv:2110.13716},
  year={2021}
}