Market-GAN

Implmentation of Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context (AAAI24)
You may download the model checkpoint and use it by putting the files into the output/ folder.

Installation

Use the AAAI_MarketGAN.yml to create a conda environment conda env create -f AAAI_MarketGAN.yml

Usage

This repo contains (1) The Market Dynamics Modeling tool, (2) Market-GAN, (3) Example usage of generated data on the downstream task. Follow these steps to walkthrogh the whole process.

I) Market Dynamics Modeling

In the MarketDynamicsModeling folder

1. Download DJI dataset from Yahoo Finance

python get_DJI_dataset.py

2. Run market dynamics modeling

(1) Move the DJI_data.csv to the data folder of Market_GAN_AAAI/data
(2) Change the "process_datafile_path=" in the config file MarketDynamicsModeling/configs/market_dynamics_modeling/djia.py to the path of the data set you downloaded
(3) Run the following command to label the data set. (the file is at configs/market_dynamics_modeling/djia.py)

use python tools/market_dynamics_labeling/run.py ( we have already done the (1) and (2) for you)

or

use python run.py --config {path of you djia.py config file} if you want to implement (1) and (2) by yourself

II) Train Market-GAN

In the Market_GAN_AAAI folder

1. Pre-train condition supervisors

sh Pretrain_DJI_V2_50.sh

2. Train Market-GAN

sh DJI_V2_RT_train.sh

3. Evaluate Market-GAN

sh Evaluate_DJI_V2_RT.sh

4. Visualize the results

sh Plot_DJI_V2_RT.sh

The results will be saved to the Market_GAN_AAAI/output. You may download the model checkpoint and use it by putting the files into the output/ folder.

III) Use Market-GAN to generate data for the downstream task

1.Generate data for the downstream tasks

(1) Export the model information for inference by running sh DJI_V2_RT_info_export.sh
(2) Generate the data by running sh run_MarketGAN.sh in the /service folder
(3) Move the generated data to downstream_tasks/data/downstream_tasks/data
(4) Run scripts for different forecasting models like sh run_MarketGAN.sh in downstream_tasks/

2.Prepare orginal data for the downstream tasks

(1) Run sh DJI.sh in downstream_tasks/data/
(2) Run scripts for different forecasting models like sh run_real.sh in downstream_tasks/

Reference

We thank these projects in helping the development of Market-GAN:
Codebase for "Time-series Generative Adversarial Networks (TimeGAN)"
timegan-pytorch
Time Series Library (TSlib)