A Study Of The Model And Algorithms Of Multi-Period Portfolio Selection

Portfolio is an important research direction in the modern financial field and the typical portfolio models are single-period and static, while the investor's behavior is often multi-period and dynamic. Therefore, the models of the multi-period portfolio selection have been studied widely in recent years. The master's thesis mainly discusses the efficient frontier of the mean-VaR model for multi-period portfolio and the model and the algorithm of the uncertainty multi-period portfolio selection, and the main contents are as follows:In the first part, it mainly introduces the researching significance and the current situation of this topic, as well as the main results of this article.In the second part, the effective solutions about mean-VaR model for multi-period portfolio selection are given, and efficient frontier is discussed, and related conclusions of the mean-VaR model for single-period portfolio selection are generalized.In the third part, firstly, because the fuzziness can express the uncertainty of the future earning, according to the credibility safety standards, a model of the multi-period fuzzy portfolio selection is established based on the credibility theory. Secondly, according to the characteristics of the model, a hybrid intelligent algorithm is designed, which is based on the theories of fuzzy simulation, genetic algorithm, wavelet neural networks and dynamic programming. The implemented steps and procedures about the algorithm are given with C programming. And empirical analysis shows that when we select the appropriate number of neurons in the hidden layer and initialization parameters about the scale, the movement of wavelet neural network, we can obtain the solution of the model, so that the portfolio behavior of the fuzzy uncertainty is operable.In the fourth part, as the fuzzy entropy is a kind of uncertain measure, another model of multi-period fuzzy portfolio selection is established based on the fuzzy entropy, and a hybrid intelligent algorithm based on the wavelet neural network is designed to solve this model and empirical analysis is implemented. The empirical analysis shows that this model is reasonable and can reflect the investor's behaviors. And the algorithm shares the better convergence and the higher computational efficiency.