Stock Trading Optimization in Emerging Market of Nigeria Using Deep Reinforcement Machine Learning and Equity Market Neutral Strategy

Investment managers are continuously looking for new technologies and better ideas to enhance their results of higher returns on investments in stock trading, and with the successes machine learning methods have been recording in other fields, it has been attracting the full attention of industry players. However, in traditional and generic portfolio strategies, forecasting of future stock prices as model inputs is a requirement, but it is not a trivial task since those values are not easy to obtain in the real-world applications. In an attempt to overcome this shortcoming, Mu-En Wu et al., (2021) developed an equity market neutral (EMN)-based portfolio management system (PMS), using reinforcement machine learning (RML), in which deep deterministic policy gradient (DDPG) and vanilla policy gradient (VPG) algorithms were used to train two neural networks; convolutional neural network (CNN) and recurrent neural network (RNN), and novel reward functions involving average returns (Aret) and Sharp ratio (Shar) were used to evaluate the performance of the developed model. Experimental results of this strategy show that the sharp ratio reward function (PMS_CNN_Shar and PMS_RNN_Shar) outperformed the returns reward function (PMS_CNN_Aret and PMS_RNN_Aret) in terms of profitability and risk. Sharp ratio reward function achieved 11.152% higher profits than those obtained using the return reward function. This study aims to improve on the study done by Mu- En Wu et al., by optimizing features selection process using ant colony optimization (ACO), the results of which show marginal improvement on the ones obtained in literature. The study also show that there is no difference whatsoever between results obtained using data from the emerging market of Nigeria and that obtained from advanced economy used in literature