Research On Portfolio Optimization Based On Reinforcement Learning Particle Swarm Optimization

Investors are often risk-averse in the investment process and hope to obtain the highest possible return.However,investment is accompanied by risks,and investments with higher returns tend to have higher risks,such as stock investment.Lack of investment concepts or improper investment methods will result in the loss of investors’ principal.Portfolio optimization refers to building a portfolio and using optimization algorithms to help investors obtain the optimal asset allocation strategy for investing in various investment products,so as to disperse investment risks and maximize investment returns.However,as the scale of investment products continues to increase,the objective function to be optimized becomes more and more complex,and the difficulty of solving the portfolio optimization problem gradually increases,which puts forward higher requirements for the performance of the optimization algorithm.As a biological heuristic algorithm,particle swarm optimization algorithm has the advantages of easy parameter adjustment and relatively easy improvement.It is widely used in solving optimization problems and has achieved rich results in many research fields.This paper proposes an improved reinforcement learning particle swarm optimizer for portfolio optimization problem,which effectively improves the return of portfolio.The main work of this paper includes:(1)Aiming at the difference of performance requirements in different stages of particle swarm optimization algorithm,an improved phased PSO algorithm(IPSO)is proposed to solve complex portfolio optimization problems.IPSO divides the optimization process into two stages.In the early stage of iteration,the nonlinear inertia weight and learning factor are used to balance the global search and local exploitation,and the contraction factor is introduced in the later stage of iteration to improve the convergence performance of the algorithm.The results show that IPSO algorithm can effectively improves the ability of PSO to solve complex portfolio optimization problems.(2)Aiming at the problem that it is difficult to update the parameters of PSO algorithm manually according to the actual operation of the algorithm,a policy gradient based particle swarm optimizer(PG-PSO)is proposed.PG-PSO constructs a policy neural network for interactive learning with the particle swarm to dynamically control the update of key parameters and improve the intelligence of the algorithm.The results of comparative experiments show that the optimization performance of PG-PSO algorithm is better than that of several improved PSO algorithms with manually configured algorithm parameters and several other optimization algorithms.(3)Learning from the idea of phased strategy,an improved PG-PSO algorithm(IPG-PSO)is proposed.The improved algorithm divides the optimization process into two stages and the adaptive updating of parameters is realized respectively through the interaction with the policy neural network,so as to further improve the optimization performance.(4)Using stock data of the Shanghai Stock Exchange 50 index and the Shenzhen Stock Exchange 100 index,this paper constructs a portfolio covering a wide range of fields.PG-PSO and IPG-PSO algorithms are used to optimize the portfolio model based on Sharpe ratio.The optimal asset allocation strategy are obtained successfully.