Reinforcement Learning for Systematic FX Trading

We explore online inductive transfer learning, with a feature representation from radial basis function network formed of Gaussian mixture model hidden processing units to direct, recurrent reinforcement learning agent. This agent is put work in an experiment, trading the major spot market currency pairs, where we accurately account for transaction and funding costs. These sources profit loss, including price trends that occur markets, are made available via quadratic utility, who learns target position directly. improve upon earlier by targeting risk context. Our achieves annualised portfolio information ratio 0.52 compound return 9.3\%, net execution cost, over 7-year test set; this despite forcing trade at close day 5 pm EST when costs statistically most expensive.