Research On Distribution Of Stock Based On Mixed Weighted Support Vector Machine

Stock market is an important means of corporate financing and stock investment.Stock market prediction research has great theoretical and practical significance for investors,enterprises and government policy making.Compared with the single day forecast,the prediction of short term trend,especially the prediction of short-term trend inflection point,has great guiding significance for both short-term investors and medium and long term investors.But the stock market is affected by real time factors,and these factors are uncertain,so the distribution of stock trend data is variable.The traditional method is to use large sample data learning model,ignoring the variability of data distribution,and it is difficult to predict the stock market inflection point effectively.Some scholars suggest sliding window method for training short-term data,which is a small sample learning problem,but it is prone to overfitting,resulting in the generalization ability of the model is not strong.Based on this,the research work of this paper is as follows:For small samples,based on knowledge gained from data distribution change detection and large sample learning,a hybrid weighted Mixed Weighted Support Vector Machine(MW-SVM)is applied to predict stock market inflection point.The new model in the weighted support vector machine model(W-SVM)is introduced in the balance to avoid overfitting,the balance is large sample learning model parameters and using the current small sample of learning the model parameters of mean square error,and using KL divergence measure between small sample distribution and large sample distribution distance.Based on the new model,this paper proposed a feature fusion energy inflection point prediction algorithm;inflection point extraction and related technical indicators,technical indicators and quantitative calculation,and then use the Markov blanket feature fusion energy of the quantitative indicators,the energy information into the MW-SVM model based on energy(MW-SVM based on energy.EMW-SVM),to predict the stock market inflection point.The experimental results show that the EMW-SVM algorithm has good performance.Based on the EMW-SVM algorithm,the stock is used for portfolio selection from the 300 shares of the Shanghai and Shenzhen stock market.The optimal portfolio strategy of stock is also a difficult problem in the actual operation of stock.At present,the combination model using historical data,ignoring the change of the recent data distribution,causes the combination to be difficult to adapt effectively,and the performance is not stable enough.This paper presents a Probability Portfolio Model(PPM)model with variable data distribution.First,the Shanghai and Shenzhen 300 stock index futures are short to hedge the big market,and the probability of winning the Shanghai and Shenzhen 300 index by the combination of the maximization of the time is the best.In return the data obey the assumed IID,using recent data simulation(revenue minus balance benefit combination of Shanghai and Shenzhen 300 index rose)probability distribution,and reference value at risk(Value at,Risk,VaR)thought,minimum income obtained under the given confidence level,and maximize the minimum income.On this basis,the KL scatter detection data distribution changes are introduced.When the KL divergence exceeds the given threshold,the model is re calculated to adapt the model to the current distribution.Finally,the numerical example of the stock market shows that the PPM algorithm has good performance.