Portfolio management system in equity market neutral using reinforcement learning

Abstract Portfolio management involves position sizing and resource allocation. Traditional generic portfolio strategies require forecasting of future stock prices as model inputs, which is not a trivial task since those values are difficult to obtain in the real-world applications. To overcome above limitations provide better solution for management, we developed Management System (PMS) using reinforcement learning with two neural networks (CNN RNN). A novel reward function involving Sharpe ratios also proposed evaluate performance systems. Experimental results indicate that PMS ratio exhibits outstanding performance, increasing return by 39.0% decreasing drawdown 13.7% on average compared trading return. In addition, more suitable construction portfolio, but has 1.98 times risk than . Among conducted datasets, outperforms benchmark TW50 traditional stocks, inferior strategy financial dataset. The profitable, effective, offers lower investment among almost all datasets. enhances well supports resource-allocation empirical trading.