Application Of Improved Artificial Fish Swarm Algorithm In Constraint Constrained Portfolio Optimization Problem

With the accelerating process of economic globalization, profound changes have happened in the environment of investment, opportunities appeared in front of the investors growing up, the investment risk along with aggravating more and more on the same time. By building effective portfolio models and data analyzing, that is vital for investors to get a target portfolio from a large number of candidate assets which could increase revenue and avoid risk to the largest extent itself. however, it is difficult to access to the relevant information accurately for investors to get effective portfolio from the unknown candidate assets, at the same time, data of the related assets to be analyzed is large. Modeling the candidate assets by improving on upon the mean-variance model which Markowitz proposed in 1952 combined with intelligent algorithm technology has become the most popular research methods, which has got great achievements in recent years.This paper intends to solve the portfolio optimization problem with cardinality constraints using an improved artificial fish swarm algorithm. Currently, most swarm intelligence algorithms typically have issues such as slow convergence speed, low convergence accuracy and weak robustness when applied to portfolio optimization problems. To this end, it proposes an improved Artificial Fish Swarm Algorithm (IAFSA) for portfolio optimization problems which tries to use standardization, put variable factors into random selection behavior and improve the search strategy of foraging behavior, and finally select the better fishes into the next generation by roulette based on the analysis of the deficiency of the previous algorithms. In addition, in order to make the model more conform to the trading environment in reality, we add the constraint conditions include increased the cardinality constraint to reduce the management cost and transaction cost in stock trading when establishing the mathematical model.In order to verify the performance of proposed algorithm, the experimental data including 20 stocks from Shanghai and Shenzhen stock markets is randomly extracted from real-time trading datasets. By analyzing parameters in both the portfolio model and the algorithm in details, experimental results prove the performance of proposed algorithm.