Improved Particle Swarm Algorithm Solve The Mean-CVaR Model Portfolio

Portfolio problem is one of the hotspot important research issues of modern finance, the main problem is minimize the risk under the premise of a given return that assign a certain amount of money to a variety of assets. In recent years, the putted forward risk measurement theory based on VaR (Value at Risk) and CVaR (Conditional Value at Risk) has been the main way to solve problem of modern financial risk investment. On the other hand, the transaction costs and the portfolio ratio upper limit has become an important constraint factor in portfolio, it is likely to lead to invalid portfolio when ignoring the transaction costs and the portfolio ratio upper limit.Therefore, this article build the portfolio model which use CVaR measure risk, consider the transaction cost and restricted stock proportion.Particle Swarm Optimization have high speed convergence and high precision, its efficient performance can be reflects in solving continuity optimization problems. But the algorithm itself exists some shortcomings like slow convergence speed later, accuracy is not high, easy to diverge, and can be seldom used to solve discrete issues. Particle swarm optimization(PSO) has a strong capability of global search, but it easily skip to global extreme, and it can only solve the continue problem, in order to overcome these disadvantages, the paper presents discrete complex method of local search, in order to improve the search capability of algorithm solve the discrete problem. PSO is easy to fall into local minimum, the paper introduce the adaptive particle migration operation to ensure the diversity of particles, avoid falling into local convergence effectively. Further enhance the convergence ability and the search speed of the algorithm.Improved particle swarm optimization in the fourth chapter apply to the mean-CVaR portfolio model build in the chapter Ⅲ. By selecting 15 stocks on Shanghai and shenzhen stock market to experiment the established model and the improved algorithm, through the selection of different parameters to study the influence on the result of the model. And the experimental results verify the effectiveness of the algorithm. Compared with other algorithm, the improved particle swarm algorithm has higher precision and more stable performance.