We propose a Reinforcement Learning (RL) algorithm for generating a trading strategy in a realistic setting, that includes transaction costs and factors driving the asset dynamics. We benchmark our algorithm against the analytical optimal solution, available when factors are linear and...