This repository contains the implementation of data augmentation for financial time series classification.

Paper Link: https://arxiv.org/abs/2010.15111

The dataset corresponds to the returns of the S&P500 constituent stocks from 1990 to 2019. It's a private dataset so it can't be shared, but in the folder raw_data there is a small example of the dataset with the stock labels randomized.

Create environment using the requirements.txt file. The code was developed in Python 3.6 and requires PyTorch==1.7.1 and tsaug==0.2.1.

$ conda create --name <env> --file requirements.txt

To process the data, run the following command. It takes a csv file with stocks returns located in the raw_data directory and builds the train and test splits of data and writes them on a new directory.

$ python3 process_data.py

Example command to run the code.

$ python3 main.py --run_path run/timewarp --da_method timewarp --hidden_dim 25 --batch_size 128 --n_epochs 200

This will create a folder <dest_dir> with the output of the run which consists of a folder output and inside, one folder per split of data containing the saved model, the evaluation metrics and the predicted best top/bottom 10 stocks to build the portfolios.

  --run_path RUN_PATH   experiment directoy (default: ./)
  --batch_size BATCH_SIZE
                        batch size (default: 256)
  --hidden_dim HIDDEN_DIM
                        hidden dimension of LSTM (default: 25)
  --n_layers N_LAYERS   number of layers in the LSTM (default: 1)
  --n_epochs N_EPOCHS   number of epochs for training (default: 600)
  --init_sp INIT_SP     initial data split (default: 0)
  --end_sp END_SP       final data split (default: 1)
  --patience PATIENCE   patience for early stopping (default: 10)
  --da_method {convolve,pool,jitter,quantize,reverse,timewarp,spawner,scaling,magnitude_warp,window_warp,None}
                        augmentation methods (default: None)

E. Fons and P. Dawson and X. Zeng and J. Keane and A. Iosifidis, "Evaluating data augmentation for financial time series classification", arXiv, 2020.

@article{fons2020dafin,
title={Evaluating data augmentation for financial time series classification}, 
author={Elizabeth Fons and Paula Dawson and Xiao-jun Zeng and John Keane and Alexandros Iosifidis},
year={2020},
eprint={2010.15111},
archivePrefix={arXiv},
 primaryClass={q-fin.ST}
}