Skip to content

Latest commit

 

History

History
106 lines (73 loc) · 5.02 KB

README.md

File metadata and controls

106 lines (73 loc) · 5.02 KB
Raw

Optimizing FOREX trading strategies with natured-inspired machine learning

The goal of the project is to optimize trading strategies based on Directional Changes using nature inspired optimization algorithms.

Algorithms used:

  • Particle Swarm Optimization (PSO)
  • A custom algo based on shuffled frog leaping - Continuous Shuffled Frog Leaping (CSFLA)

To do so, I use the trading strategy provided by [1] which use Genetic Algorithms to find a suitable set of parameters for a Directional Change - based strategy.

The problem can be resumed to optimizing a fitness function - which is the performance of the trading strategy given a set of parameters.

To ensure robustness of my proposed algorithms, I test them with the same configuration that the authors in [1].

Contents

This repository contains:

  1. Custom, from scratch implementations of the PSO and CSFLA algorithms, in the .py files in the root of the repository
  2. 12 months of 10-min FOREX data on 4 currency pairs, in the data/ folder, used to train and test the algorithms
  3. Configuration files, in the config/ folder, which are used by the program to generate experiments
  4. Analysis notebooks in the analysis/ folder that cover - the algorithm parameter tuning process - analysis of the trading strategy performance on the test data
  5. Results of all the experimented and final strategies, in the results.zip file
  6. Project documentation, including UML class diagrams and meeting powerpoints, in the docs/ folder

The experiment:

Installing required packages:

  1. Recommended installation: 
# Make sure to replace <envname> with the name of your env
conda create --name <envname> --file requirements.txt

On Windows, in your Anaconda Prompt, run activate <envname>

On macOS and Linux, in your Terminal Window, run source activate <envname>

OR Installation with pip (less recommended)

  • Install the required libraries with pip:
    pip install requirements.txt
  1. Install an Ipython notebook reader (provided by Anaconda CLI )

Running a short demo:

If you simply to test out the system, you can run the simplified (and thus poorly performing!) algorithms on one month of data for one currency pair:

    python main.py -c ./config/demo.json

Running the full experiment (optional):

The full experiment generates all the configurations results (51 PSO configurations, 41 CSFLA configurations) on test and training data. This can take up to several days if your machine has low computational power.

Pre-computed results can be extracted from the given results.zip file:

    unzip results.zip

Otherwise, to run the full experiment:

On the training data:

    python main.py -c ./config/training_config.json -p config/algos/pso_param_exp_configs.json -f config/algos/csfla_param_exp_configs.json -o
    python main.py -c ./config/training_config.json -p config/algos/pso_configs.json -f config/algos/csfla_params_config_2.json -o
    python main.py -c ./config/training_config.json -p config/algos/pso_configs_2.json -f config/algos/csfla_configs.json -o
    python main.py -c ./config/training_config.json -p config/algos/pso_configs_3.json -o

On the testing data:

    python main.py -c ./config/testing_config.json -p config/algos/pso.json -f config/algos/csfla.json -g

Analysing the results

Static notebooks:

To read the notebooks without making changes and having to get the data, you can open the .html files in the analysis/ folder.

The test_data_analysis.html file presents an analysis of the results of the algorithms on the test

The preliminary_pso_tuning_analysis.html and indepth_pso_tuning.html files present parameter tuning analysis of the PSO.

The preliminary_csfla_tuning_analysis.html and indepth_csfla_tuning.html files present parameter tuning analysis of the CSFLA.

To run the interactive notebooks:

Make sure you have either extracted the provided results, or generated the training data.

Run the Ipython notebooks in the analysis/ folder.

The test_data_analysis.ipynb notebook presents an interactive analysis of the results of the algorithms on the test

The preliminary_pso_tuning_analysis.ipynb and indepth_pso_tuning.ipynb notebooks present interactive parameter tuning analysis of the PSO.

The preliminary_csfla_tuning_analysis.ipynb and indepth_csfla_tuning.ipynb notebooks present interactive parameter tuning analysis of the CSFLA.

References:

[1] - M. Kampouridis and F. E. B. Otero, "Evolving trading strategies using directional changes," Expert Systems with Applications, vol. 73, pp. 145-160, 2017.