Financial Trading via Reinforcement Learning and Convolutional Neural Networks

Abstract

The paper presents a novel approach to developing a pair trading strategy for cryptocurrencies by employing a customized Rainbow DQN and an image encoding technique. This method transforms candlestick features of Bitcoin (BTC) and Ethereum (ETH) into images, which are then input into a Convolutional Neural Network (CNN) for feature extraction before being fed into the Rainbow DQN model for making trading decisions. Although the Rainbow DQN strategy outperforms RSI and correlation strategies, none of them are profitable after factoring in transaction costs.