Deep reinforcement learning for portfolio management
In our paper, we apply deep reinforcement learning approach to optimize investment decisions in portfolio management. We make several innovations, such as adding short mechanism and designing an arbitrage mechanism, and applied our model to make decision optimization for several randomly selected portfolios. The experimental results show that our model is able to optimize investment decisions and has the ability to obtain excess return in stock market, and the optimized agent maintains the asset weights at fixed value throughout the trading periods and trades at a very low transaction cost rate. In addition, we redesigned the formula for calculating portfolio asset weights in continuous trading process which can make leverage trading, that fills the theoretical gap in the calculation of portfolio weights when shorting.
PDF Abstract
