joseph-mcallister/quant-playground

quant-trading

Introduction

This is a collection of me playing around with crypto data and building out some basic ML strategies.

Data Sources

• Bitcoin Coinbase Exchange Historic Prices: https://www.coinbase.com/api/v2/assets/prices/5b71fc48-3dd3-540c-809b-f8c94d0e68b5?base=USD
• Coinbase API Docs: https://docs.pro.coinbase.com/#get-historic-rates

Potential Features

• Last 24 hour spot price
• Last n days daily Bitcoin price data
• Google Search Trends Previous 7 days
  • "is bitcoin crashing"
  • "how to buy bitcoin"
• Last n days daily S&P price data

Technical Indicators

• Bollinger Bands
• Relative Strength Index (RSI)
• Momentum
• Z-score

Setup

1. Install Python3 and pipenv
2. pipenv shell to start a shell with the virtual env
3. pipenv install to install dependencies to the virtual env
4. pip install git+https://github.com/bartosh/backtrader.git@ccxt install Ed Baratosh's branch with cctx support (never added to main repository)
5. python -m ipykernel install --user --name=quant-trading to make virtual env available as kernel for jupyter
6. jupyter notebook

Backtrader

Requires a special branch of backtrader with support for cctx (live exchanges) for data feeds

Baseline

Scenario

• $10,000 starting cash
• January 1st, 2016 - December 31st, 2020
• Strategy: Market Buy and HODL Performance
• Total Return: 413%
• Sharpe Ratio: 0.707
• Maximum Drawdown: 83.69361893711935%

Neural Network Strategy Price Only

Model

• Neural Network binary classifier on hourly tick data
• Features are all moving average diffs over different lookbacks Performance

Plot