Hero Background

WELCOME

Clarence Crypto Trader is a project initiated by a group of (relatively) humble data scientists based in Sweden and Switzerland. The main objective of the project is to develop, test, and deploy a binary-classification-based machine learning model with the aim of performing automated live trading simulations, all with the use of Google Cloud Platform (GCP) components.

The project has been in the works since July of 2024. Many challenges were encountered and solutions implemented. However, there is still much to be accomplished before live trading simulations can be performed and the trade outcomes displayed, namely the steps mentioned here:

  1. Complete backtesting on the best-performing model, record results and fine-tune the training process to enhance model precision

    Key components: Google Compute Engine (GCE), Sklearn, Python
  2. Develop an informed trading strategy for the simulations, including but not limited to stop-loss and risk analysis strategies

    Key components: Financial Analysis, Python
  3. Configure VM architectures to reliably run the simulation using Google Compute Engine and storage solutions to store automatic trade operations and trade outcomes, to be displayed on the site

    Key components: Google Compute Engine (GCE), BigQuery, Google Cloud storage

July 2024

A literature review was conducted, and historical market data from different crypto exchanges were analysed

Scrum was chosen as the Agile development methodology

Market data transformation libraries for Python and Javascript where chosen

Sklearn's RandomForestClassifier was chosen as the initial model to be trained

Google Colab was used for this duration

August 2024

XGBoost and a Fully Connected Neural Network (FCN) were tested, both with inferior validation test results

Compatibility issues identified when transitioning to AWS, GCP was opted for instead

Google Colab together with GCP and AWS tools were used for this duration

September 2024

Incompatibility between the Binance exchange API and GCP components led to Kucoin being used as the exchange instead

Re-thinking of the strategy and approach due to the change in exchange platform, data anlysis of Kucoin historical data ensues

Trained RandomForestClassifier models on Kucoin historical data and stored candidate models

October 2024

ETL pipeline implemented using Google Cloud Run, storing data in Google Cloud Storage

Backtesting of best-performing models begins, with live data from the Kucoin API

November 2024

The project is currently on hold as we reassess our strategy and objectives.

Further evaluations will determine the next steps for the trading simulations.