The Application Of Machine Learning Models To A Share

The financial market plays an important role in modern economy.On the one hand,for the policymakers,the effective forecasting of the financial market can make them better predict the macro economy and then specify relevant policies to guide the healthy operation of the economy.On the other hand,for financial market participants,the financial market forecasting is particularly important for portfolio construction and risk management.It is difficult for traditional models,e.g.ARMA,to capture nonlinear information,and the prediction results are often not very ideal.Since the 1990s,as a branch of computer science,the concept of artificial intelligence(AI)has been constantly mentioned.Academia,industry and commerce,and even the general public are all enthusiastic about AI.With the improvement of computer performance and the optimization of algorithm,the theory,method and technology of AI have been generally valued and improved.Especially in the fields of image recognition,speech recognition.automatic driving and machine translation,etc.,AI has gradually moved from behind the scenes to the front stage,and started to expand its application scope in many related industries,which can not help but remind people of the bright prospect of AI application in financial markets.In this paper,machine learning model is applied to Chinese A-share market,mainly using support vector machine(SVM)model,artificial neural network model,random forest model and the integration model to predict the direction of Chinese A-share stock the next three days.According to the probability of rise and fall,M stocks with the greatest probability are selected to construct a portfolio,including acquiring data,the construction of feature variables and target variables,the split of training set and test set.data preprocessing,feature selection,model parameter turning,model evaluation and other processes.The results show that the integrated model based on vector machine model,artificial neural network model and random forest model performs best Among them,when M is 5,the performance is the best,Sharp ratio is 1.35,annual compound return is 44.69%,and the performance of the integrated model is significantly better than Hu-shen 300 index and Zhongzheng 500 index.In addition,the industry distribution of strategy portfolio is also different to some extent,which is not an equilibrium distribution.It indicates that the model can capture the information of A-share industry rotation to some extent,and provides A new idea for quantitative research and quantitative stock selectionCompared with other literatures,using a single model,or synthesizing other feature selection models on the basis of a single model,the innovation of this paper lies in creatively integrating support vector machine(SVM)models,artificial neural network models and random forest models with low correlation and different inherent logical and Mathematical Foundations Based on predictive probability.The results show that the integration is simple.The performance of the latter model is better than that of any single model,which embodies the advantages of model integration.What's more,it enriches the data dimension.On the basis of simple price data and technical indicators,it adds some daily data,including valuation data,turnover rate data,volatility data and momentum reversal data,etc.,so as to make the data dimension more diversified and reflect more sufficient information,thereby improving the accuracy of model prediction and the yield of the investment portfolio constructed accordingly.These are more in line with the nonlinear and nonstationary nature of the financial market and can produce better application effects in prediction and constructing investment strategies.