Stock Forecasting System Based On LSTM Analysis Of Hong Kong Position Factors

Since the Shanghai-Hong Kong stock connect program and Shenzhen-Hong Kong stock connect program started in 2014 and 2016,with the continuous improvement of financial market openness,more and more international capital(referred to as Hong Kong capital in this thesis)flows into the A-share market through the Hong Kong Stock Exchange.By the end of April 2020,the stock market value of A shares held by Hong Kong capital has reached 1.5 trillion yuan,and Hong Kong capital is exerting more and more influence on A stock market.Hong Kong Stock Exchange makes public its detailed changes of Hong Kong capital position data after everyday trading,which makes more and more investment institutions devote themselves to the in-depth mining and analysis of Hong Kong capital position data,in order to find the law between the flow of Hong Kong capital and the changes of stock price,and then provide guidance and reference for investment strategy,so as to reduce investment risk and improve the return of investment strategy.However,the performance of many of the data analysis models for Hong Kong investment is not good enough for actual needs.Thanks to the extensive application of artificial intelligence technology in the financial field in recent years,in order to further mine and analyze the change rule between the flow direction of Hong Kong capital and the stock price,in this thesis,the Long Short-Term Memory network in the deep learning algorithm is introduced into the data analysis model of Hong Kong capital to mine and analyze a large number of Hong Kong capital transaction data.Good experimental results are obtained in this thesis.The main contents and contributions of this thesis are as follows:1.Data pretreatment and factor selection.Because the existing dataset is not suitable for our research,we grab more than 1 million pieces of data of Hong Kong capital positions published by the Hong Kong stock exchange from June 29,2016 to January 23,2020,and combine them with more than 500 stock price factors recognized in the field of quantitative investment to form a data set of Hong Kong capital positions,and clean and configure the data.Aiming at the problem that the number of factors is large and the influence of factors on the stock prices is unknown,the random forest is used to mine,analyze and filter the factors to find out the factor set which has a great influence on the stock price.2.Propose a Long Short-Term Memory(LSTM)based stock price volatility prediction algorithm.Combined with the characteristics of the data set used in this thesis,the multi factor LSTM is used as the basis,and the model is improved,and attention mechanism(attention mechanism)is added Mechanism)innovatively put forward the integration of self attention and multi factor LSTM model to predict the volatility of stock price,which improves the accuracy of model prediction.Experimental data show that the algorithm has better prediction accuracy and more general applicability.3.Build a stock recommendation system based on the price trend prediction of stock using above-mentioned methods.The data used to train the model will be updated everyday.Base on the prediction results of the fusion model based on self attention and multi factor LSTM,We will recommend the stocks with high expected return to users.