Αλγόριθμοι αυτοματοποιημένων συναλλαγών βασισμένοι σε τεχνικές μηχανικής μάθησης
Machine learning-based automated trading algorithms

View/ Open
Keywords
NEAT ; Reinforcement learning ; Machine learning ; TradingAbstract
One relatively new approach in crypto currency trading is to use machine learning algorithms to predict the rise and fall of prices before they occur. An optimal crypto currency trader would buy before the price rises and sell before the value declines.
This thesis presents an approach combining a recurrent reinforcement learning method (RRL) model with the NeuroEvolution of Augmenting Topologies (NEAT) to generate a trading signal – neural network capable of achieving high returns in crypto currency with low associated risk. Our goal is to feed our network with an input vector each time using the previous output of our neural network. To create our input vector, we are using a time series model of investment returns, the position of our trader which is short, long, and neutral plus a weighted factor which is equal with one.
The proposed approach has been tested with real daily data of daily cryptocurrency data, where we have selected the well-known cryptocurrency BTC (Bitcoin). Sharpe ratio is the metric that we have used in our Project to evaluate our model. The above-mentioned algorithm attempts to maximize Sharpe ratio. The results achieved show that Sharpe ratio increases within training period therefore we are getting good results also during testing period.