Algorithmic trading on cryptocurrency markets using machine learning techniques

Abstract

The aim of this thesis is to apply different machine learning models to the price prediction problem in cryptocurrency exchange markets in order to find profitable trading algorithm. Eight trading pairs are taken into consideration for the thesis, and Open-High-Low-Close (OHLC) price data of those pairs are utilized through well known linear models and tree-based models. Besides, an ensemble model framework is proposed for this specific task. Trading simulations are conducted based on two different experimental design. First, each trading pair is considered separately in turn and trading agents are simulated as if they can trade only one trading pair. Then, all selected trading pairs are taken into account together and trading agents that are capable of trading any of the selected trading pairs are simulated. Model performances are presented and compared with each other and with naive buy-and-hold approach. Although statistical tests are failed to show that results are statistically significant, 22-fold return is achieved in 5-month test period with the proposed method.