2016FX Trading via Recurrent Reinforcement Learning.pdf

FX Trading via Recurrent Reinforcement Learning Carl Gold Computation and Neural Systems California Institute of Technology, 139-74 Pasadena, CA 9 1125 Email January 12,2003 Abstract: of the input, the neural networks are called “recurrent”. The This study investigates highfrequency currency trading output of the network F E [-1,1] at time t is the position with neural networks trained via Recurrent Reinforcement (long/short) to take at that time. Neutral positions are not Learning (RRL). We compare theperformance of single allowed so the trader is always in the market, also known as layer networks with networks having a hidden layel; and a “reversal system”. For a single layer neural network (also examine the impact of thefixed system parameters on known as a perceptron) the trading function is performance. In general, we conclude that the trading M systems may be effective, but thep e r f o m n c e varies widely Ft = s i g n ( x W;rt-i -I- WM+iFt-i +V) for different currency markets and this variability cannot i=O be explained by simple statistics of the markets. Also we find that the single layer network outpei$onns the two layer where lu’ and w are the weights and threshold of the n