Pairs Trading for Cryptocurrencies

Honors Capstone for Michael Ward.
Utah State University -- May 2018.

http://capstone.mikethebro.com

Abstract

This project takes several common strategies for algorithmic stock trading and tests them on the cryptocurrency market. The three strategies used are moving average crossover, mean reversion, and pairs trading. Data was collected every five minutes for the top one hundred cryptocurrencies starting October 5, 2017. Due to the high volatility of the market, the data includes various market situations. Three noted situations are a rising market, falling market, and relatively stable market. The three strategies were modified to optimally follow each market situation. Modifications include adjusting parameters used in each strategy as well as mixing several strategies or dynamically changing between strategies. In each strategy and with each cryptocurrency, the benchmark the algorithm is tested against is the market's performance, or what an investor would have after buying and holding. Returns are compared with the buying and holding strategy, and different scenarios are analyzed to determine the risk associated with buying and holding compared with an algorithmic strategy. Results will be taken with the market's actual trends and also with some alternate possible trends to test all market scenarios. A web interface will accompany the presentation, allowing users to test the strategies by entering their own parameters and instantly see the results.

Algorithms

Pairs Trading

Finding the relationship between two or more prices.

Moving Averages

Algorithms using the moving averages of the stock price over multiple days.

Mean Reversion

Improvement on Moving Averages where algorithms will place limits above and below the moving average.
This will allow buy and sell times to more closely approach the highs and lows.

Dynamic Mean Reversion

Improvement on Mean Reversion algorithm. Algorithm will dynamically change limits in accordance with stock's volatility.

Results

Moving Average Crossover on Bitcoin

With an initial investment of one Bitcoin and tested over a year and trading a maximum of once a day, the algorithm returned:
Total Profit: $1,894.86
Percent Return: 311.3%
Percent Return (buying and holding): 659.1%

With an initial investment of one Bitcoin and tested from October 5 to November 3 and trading a maximum of once every 5 minutes, the algorithm returned:
Total Profit: $6,150.09
Percent Return: 163.9%
Percent Return (buying and holding): 142.3%

**While both algorithms returned less than buying and holding, moving average crossover may be more protected in case of a market crash. Further research will follow.

Tasks

Research

• [x] Write my first algorithm
• [ ] Learn how the stock market works: optimal times to buy and sell and how to best predict future changes.
• [ ] Learn the basics of algorithmic trading.

Data Setup

• [x] Get past data
• [ ] plan to organize data
• [ ] sub-repo or regular data files?
• [ ] Create an account on QuantConnect and try trading

Pairs Trading

• [ ] Dickey-Fuller algo in Python
• [ ] Johansen cointegration test in Python (on Quantopian)

Moving Average Crossover

• [ ] Learn Moving Averages Crossover strategy.
• [ ] Test Moving Averages Crossover strategy with Bitcoin.
• [ ] Modify MVA test with Bitcoin to find better results.

Writeup

• [ ] First Thesis Draft.
• [ ] Second Thesis Draft.
• [ ] Penultimate Thesis Draft
• [ ] Final Thesis Submission
• [ ] Present Thesis at Research Symposium

Timeline

Meetings

Research Sources

• Quantopian
• QuantConnect
• Investopedia
• Robinhood Markets
• Market Watch
• Stock Trading

Final Product

The final product will consist of a written report. Sample code and graphs for the performance of each algorithm will be included in the report. All code will be displayed on this repository.