Optimization Algorithm And Empirical Analysis Of Portfolio Related Model

With the rapid growth of the financial derivatives market and the emerging Internet finance,people have led to the upsurge of investment and financial management.Individuals and enterprises have an urgent need for the rational theory of wealth distribution.This is the core content of portfolio theory.The main problem of it is how to allocate existing wealth in venture capital assets reasonably,so as to achieve the goal of maximizing revenue and maximizing cumulative returns under established risks.At present,there are two main directions of portfolio theory in finance field,one is the mean variance theory based on Markowitz,and the other is based on Kelly capital growth theory.Although the mean variance model of Markowitz a quantitative investment precedent,and excellent performance has been widely used in practice,but there are still some shortcomings: poor interaction,not for continuous investment decision-making;risk measurement is not scientific,the variance with positive deviation and negative deviation alike;micro weight too much,resulting in operation decrease and increase transaction cost.In view of the existing problems,this paper proposes a multi stage Mean-CVaR model under the weight separation constraint.First,the multi stage continuous investment is realized through the way of moving the historical data window,and the problem of poor interactivity is overcome.Then,the risk value CVaR is chosen as a risk measure with an asymmetric condition,and only the negative deviation part of the mean is considered as a risk.At the same time,the weight separative constraint is added to the two planning optimization model,and the problem of too much weight is solved by the linear mixed programming.Finally,the stock of the Shanghai 50 index is selected to carry out the simulation experiment to prove the effectiveness of the Mean-CVaR model under the weight separation constraint.With the rapid development of artificial intelligence,the combination of Kelly capital growth theory and online learning algorithm has led to the study of online investment portfolio.Compared with Markowitz's portfolio theory,online portfolio theory is more consistent with the characteristics of multi-stage,multi interaction and dynamic adjustment in the actual investment process,which is favored by investors and scholars.At present,there are a series of online portfolio strategies in academia.One of the most rapidly developing research directions is the online portfolio strategy based on reverse strategy and passive active algorithm.This kind of strategy mainly consists of two parts,the estimation of relative price and the optimization of the portfolio.The online dynamic portfolio selection(OLES)based on adaptive dynamic exponential smoothing method is proposed in this paper,which gives different stocks different smoothing coefficients,and achieves a more accurate relative price prediction through dynamic adjustment of gradient descent method.Through the simulation experiments of multiple actual data sets and the comparison of many financial indicators,the effectiveness of OLES is demonstrated.