Forecasting Stock Prices Using Stock Correlation Graph

We create a new modle by combining GCN model with GRU model and achieve around 5% incraese in accuracy in comparison to baseline model.

Published Paper

Forecasting Stock Prices Using Stock Correlation Graph: A Graph Convolutional Network Approach

Paper Citation

X. Yin, D. Yan, A. Almudaifer, S. Yan and Y. Zhou, "Forecasting Stock Prices Using Stock Correlation Graph: A Graph Convolutional Network Approach," 2021 International Joint Conference on Neural Networks (IJCNN), 2021, pp. 1-8, doi: 10.1109/IJCNN52387.2021.9533510.

Link to the Paper

https://ieeexplore.ieee.org/document/9533510

Model Architecture

Illustration of Graph Convolution

Dependencies

• Tensorflow
• Pandas
• Numpy
• Sklearn
• Configparser

Workflow

The workflow is utralize. Start training the model by simply run:

python3 train.py

Hyper parameters can be changed at the file

config.ini

Hyper Parameter

Hyper Parameter	Definition
data_addr	The address for the data(stock prices) in use
adj_addr	The address for the adjcency matrix in use
s_index	The index of stock to predict
lr	Learning rate
n_neurons	Number of neurons in GRU layer
seq_len	Sequence length
n_epochs	Number of epochs
batch_size	Batch size
th	Threshold for ε-insensitive accuracy

Data Address

The form of data is address is compose of several parts

./data/data/[dataset]/[dataset]_[time duration]_price.csv

For in stance:

./data/data/dow/dow_1day_price.csv

Adjacency Matrix Address

The form of data is address is compose of several parts

./data/adj/[dataset]/[time duration]/[dataset]_[time duration]_[cut off]_01_price.csv

For in stance:

./data/adj/dow/1day/dow_1day_090_01_corr.csv

*** [dataset] and [time duration] must be the same for both data address and adjacency matrix address

Directory Structure

gcgru_stock_prediction
├── config.ini
├── gcgru.py
├── input_data.py
├── train.py
├── utils.py
├── image
│   ├── gcc.jpeg
│   └── model.jpeg
└── data
    ├── adj
    │   ├── dow
    │   │   ├── 1day
    │   │   │   ├── dow_1day_050_01_corr.csv
    │   │   │   ├── dow_1day_055_01_corr.csv
    │   │   │   └── ...
    │   │   └── ...
    │   └── etf
    │       ├── 1day
    │       │   ├── etf_1day_050_01_corr.csv
    │       │   ├── etf_1day_055_01_corr.csv
    │       │   └── ...
    │       └── ...
    └── data
        ├── dow
        |   ├── dow_10min_price.csv
        |   ├── dow_15min_price.csv
        |   ├── dow_1day_price.csv
        |   ├── dow_1h_price.csv
        |   └── dow_30min_price.csv
        └── etf
            ├── etf_15min_price.csv
            ├── etf_1day_price.csv
            ├── etf_1h_price.csv
            └── etf_30min_price.csv