/deepOBs

The Short-Term Predictability of Returns in Order Book Markets: A Deep Learning Perspective.

BSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

deepOBs

This project aims to explore and analyze order book-driven deep learning models for predicting high-frequency returns in financial exchanges with a limit order book market structure.

The main methods, which we believe could be of use to other researchers, are found in:

  • data_process.py: functions for processing order book data dowloaded from LOBSTER to raw order book, order flow and volume features and the corresponding returns;
  • data_methods.py: auxiliary functions for processed data;
  • custom_datasets.py: create custom tf.dataset objects to load features and responses into models;
  • model.py: a class to build, train and evaluate deepLOB (Zhang et al., 2019), deepOF (Kolm et al., 2021) and deepVOL as keras.models.Model objects;
  • MCS_results.py: functions to perform the bootstrap Model Confidence Set (Hansen et al., 2011) procedure on results

alt text

Release v1.0 DOI

This release contains code and results for the paper 'The Short-Term Predictability of Returns in Order Book Markets: A Deep Learning Perspective'.