Portfolio optimization in cryptocurrencies: a comparison of deep reinforcement learning and traditional approaches

Abstract

Cryptocurrencies have become appealing investment options in recent years because of their high potential returns. This asset class emerged as a unique investment opportunity with distinguishing characteristics such as decentralized nature and uncorrelation with other assets. Investing in this product, however, has become a hazardous venture due to its extreme volatility and unpredictable price swings. As a result, a portfolio optimization is an essential tool for investors seeking to reduce risk while aiming for high returns. This thesis studies the Deep Reinforcement Learning models applied to cryptocurrency portfolio optimization compared to traditional methodologies like Markowitz's and rudimentary equally weighted portfolios.