Discovering Predictable Latent Factors for Time Series Forecasting

This repository is the official implementation of "Discovering Predictable Latent Factors for Financial Time Series Forecasting".

Optional: include a graphic explaining your approach/main result, bibtex entry, link to demos, blog posts and tutorials

Requirements

To install requirements and the Qlib toolkit:

# Install the requirements
pip install -r requirements.txt

# Install Qlib
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

Training

Ours+LR

# CSI100
python train.py --data_set csi100 --K 0

# CSI300
 python train.py --data_set csi300 --K 0

Ours+HIST

# CSI100
python train.py --data_set csi100

# CSI300
 python train.py --data_set csi300

Evaluation

To evaluate my model on ImageNet, run:

# Ours+LR
python eval.py --model_path <path_to_model> --K 0 --data_set <csi100/csi300>

# Ours+HIST
python eval.py --model_path <path_to_model> --data_set <csi100/csi300>

Contributing

The framework of our code is based on the code in "The HIST framework for stock trend forecasting" of the work "HIST: A Graph-based Framework for Stock Trend Forecasting via Mining Concept-Oriented Shared Information" by Xu et al. (WWW 2022).