Market trading has become popular among most people these days, though most of them avoid making risky action, because of lack of knowledge and the difficulty to predict its movements.
The project goal is to meet the need to ease a portfolio managing, using a system which takes actions in relative short time intervals, with a high profit potential. The system goal is to maximize the profit using US capital market’s stocks and bonds, while remaining in a specified ratio between the value of the stocks and bonds in the portfolio, to hedge risk.
The selected solution for the project was using an AI system, based on Neural Network which commits a Reinforcement Learning process of Deep Q Learning (DQN). Conduct in the portfolio is being done by an agent, who begins with amount of cash, stocks and bounds, and he must learn the set of actions which will lead him to maximize his assets in the long term.
The training data is based on real 13 years data form the US market. In the training procedure data groups are being collected: state, action, reward (change in portfolio value as action’s outcome) and next state arrived at. The action is being selected as the one that maximizes the current reward and next state’s value, as done in DQN method. Also, selling with positive profit leads to tax payment.
The results were varied, and included difference in the final profit measurement, as a result of the networks which were trained. During the project it was found that the way of choosing and using the data, the freedom the agent was given, and the number of available actions, all had a great impact on the network’s results.
Training was made on various networks, some of which achieved better results than a “take-no-action policy” (letting the portfolio gain value just from market changes). It is enough to prove the feasibility and to strengthen the need in this project.