Exploring the Use of Random Forests as a Basis for Portfolio Selection in the Crypto Market

Abstract

This study explores the use of Random Forest models for portfolio selection in the cryptocurrency market. Looking at previous work, most of the research has focused on short-term price forecasts for individual cryptocurrencies, hence the need for more extensive research into portfolio selection in the cryptocurrency market using machine learning techniques. The project's goal is to build an optimal cryptocurrency portfolio using Markowitz's portfolio theory, incorporating the predictions generated by a Random Forest model trained on technical indicators. The optimal portfolios were determined using the Sharpe ratio and used to build a trading strategy. Our analysis shows that the overall return is negative, but holding the strategy for less than a year leads to positive returns with a higher rebalancing frequency.
Furthermore, the risk-adjusted performance measured by alpha shows that the higher frequency rebalancing strategies outperform the benchmark, particularly the 2-month rebalancing approach. Furthermore, the model performs better than the benchmark index during bearish market conditions, suggesting it could be utilized as a hedging strategy to reduce portfolio risk. Further, the benefit of the addition of our crypto strategy to a classic 60/40 was examined. Still, we found that at least for the current dataset, the performance of the traditional portfolio is hindered by the addition of our crypto portfolio. That is due to the high correlation between the crypto and equity markets. Further research and analysis are necessary to fully understand the potential benefits and drawbacks and to develop sophisticated strategies and investment vehicles to optimize returns and manage risks.

Date of Award	Apr 2023
Original language	American English
Supervisor	IBRAHIM Elfadel (Supervisor)