Multiagent-Based Portfolio Simulation Using Neural Networks

摘要

The focus of this study is on creating a novel platform for portfolio simulations using neural networks, which are uniquely designed to imitate various well known investing strategies. The artificial investing agents are not only forecasting but also managing their portfolios. The proposed multi-layered framework constructs (i) an intelligent agent composed of a few trained neural networks, each specialized to make investment decisions according to the assigned investment strategy, (ii) multi-agent systems, which are trained to follow different investing strategies in order to optimize their portfolios. The novel multiagent-based portfolio simulation platform gives us an opportunity to display the multi-agent competitive system, and find the best profit-risk performing agents, i.e. investing strategies. In sum, simulations show that the proposed NN-based multi-agent system produces similar results in terms of e.g. wealth distribution compared with empirical evidence from the actual investment markets' behavior as described by Pareto wealth and Levy returns distributions.