Deep Reinforcement Learning of the Rules of Financial Assets Trading

Abstract

The foreign exchange (Forex) market is the world’s largest currency exchange market where diverse people globally make financial investments. Since the advent of this market, a leading challenge for capital investors and analysts has been devising a versatile solution and strategy to make profitable investments in this market. Recently, machine learning (ME) and, particularly, deep learning (DL) algorithms, in tandem with intelligent learning systems working based on previous data, have markedly fostered functions such as data classification or feature extraction. Likewise, the advent of reinforcement learning (RL) and intelligent systems has revolutionized learning and operating with no need for previous data and merely based on algorithms such as Q-Learning. The present research operated an intelligent agent (IA) trained based on DL and Q-Learning algorithms in RL to make financial transactions in the Forex market that deliver the greatest profitability and least losses within the test period. For several major currencies in the Forex market, this RL-based IA was compared for 24 hours with a trading strategy based on trend detection algorithms employing diverse ME classification techniques. The proposed IA delivered higher profitability values of 64.9% and 38.5% compared to the previous structures in aud/usd and eur/usd currencies, respectively. Collectively, the IA delivers higher profitability than the ME-based trading strategies.