Abstract
Cryptocurrency is a cryptography-based digital asset with extremely volatileprices. Around $70 billion worth of crypto-currency is traded daily onexchanges. Trading crypto-currency is difficult due to the inherent volatilityof the crypto-market. In this work, we want to test the hypothesis: "Cantechniques from artificial intelligence help with algorithmically tradingcryptocurrencies?". In order to address this question, we combine ReinforcementLearning (RL) with pair trading. Pair trading is a statistical arbitragetrading technique which exploits the price difference between statisticallycorrelated assets. We train reinforcement learners to determine when and how totrade pairs of cryptocurrencies. We develop new reward shaping andobservation/action spaces for reinforcement learning. We performed experimentswith the developed reinforcement learner on pairs of BTC-GBP and BTC-EUR dataseparated by 1-minute intervals (n = 263,520). The traditional non-RL pairtrading technique achieved an annualised profit of 8.33%, while the proposedRL-based pair trading technique achieved annualised profits from 9.94% -31.53%, depending upon the RL learner. Our results show that RL cansignificantly outperform manual and traditional pair trading techniques whenapplied to volatile markets such as cryptocurrencies.