Finformer: Forecasting Financial Factors with Informer

For CSCI 4900: Deep Learning

This repository contains a modified implementation of the Informer model, used to experiment with the long sequence time-series prediction of financial factor data.

Goals

• Apply Informer model to financial time-series data (Fama-French factors)
• Benchmark performance against persistent and zero baselines
• Identify which factors exhibit the most predictable structure
• Evaluate multistep forecasting with a variety of architectural modifications, and hyperparameter settings

Getting Started

1. Clone the repository

git clone https://github.com/mmocklin18/finformer.git
cd finformer

2. Create a virtual environment

python3 -m venv .venv
source .venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Run an experiment

python -u main_informer.py \
  --model informer \
  --data custom \
  --root_path ./data/ \
  --data_path factors_data.csv \
  --features S \
  --target UMD \
  --freq d \
  --seq_len 60 \
  --label_len 30 \
  --pred_len 10 \
  --enc_in 1 \
  --dec_in 1 \
  --c_out 1 \
  --e_layers 2 \
  --d_layers 1 \
  --n_heads 4 \
  --d_model 128 \
  --d_ff 256 \
  --dropout 0.1 \
  --train_epochs 10 \
  --batch_size 16 \
  --learning_rate 0.001 \
  --loss mse \
  --gpu 2 \
  --itr 5

For a complete list of command-line arguments and configuration options, refer to the original Informer repository

References

@article{haoyietal-informerEx-2023,
  author    = {Haoyi Zhou and
               Jianxin Li and
               Shanghang Zhang and
               Shuai Zhang and
               Mengyi Yan and
               Hui Xiong},
  title     = {Expanding the prediction capacity in long sequence time-series forecasting},
  journal   = {Artificial Intelligence},
  volume    = {318},
  pages     = {103886},
  issn      = {0004-3702},
  year      = {2023},
}

@inproceedings{haoyietal-informer-2021,
  author    = {Haoyi Zhou and
               Shanghang Zhang and
               Jieqi Peng and
               Shuai Zhang and
               Jianxin Li and
               Hui Xiong and
               Wancai Zhang},
  title     = {Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting},
  booktitle = {The Thirty-Fifth {AAAI} Conference on Artificial Intelligence, {AAAI} 2021, Virtual Conference},
  volume    = {35},
  number    = {12},
  pages     = {11106--11115},
  publisher = {{AAAI} Press},
  year      = {2021},
}

This project builds on the original Informer implementation by Zhou et al. (AAAI 2021), used under the MIT License.