Research On Stock Price Trend Prediction Based On Multi-input Deep Learning Method

Financial market is an important driving force of a country’s economic development.Multi-level capital market plays an important role in supporting development of real economy.As China’s capital market reform and opening-up continues to accelerate,China’s stock market is increasingly favored by domestic and foreign investors.However,stock market is unstable and may bring losses to investors.In order to deal with market risks,it is necessary to establish stock price prediction model.Because stock prices are affected by a variety of uncertain factors such as macro economy,political events and investors’ expectations,stock price data presents nonlinear and non-stationary characteristics with low signal-to-noise ratio.Therefore,traditional linear methods have great limitations in prediction problems.Deep learning effectively solves these problems.It has advantages of robustness,self-learning ability and strong fitting ability.Deep learning can effectively improve prediction accuracy and has been widely used in the field of stock price prediction.On the basis of LSTM(Long-Short-Term Memory),this paper proposes Multi-input LSTM,which is combined with wavelet transform to predict the trend of SSE Composite Index.It can not only mine valid data in time series,but also support different types of data as input,which can effectively improve prediction accuracy.The whole model is divided into two stages.In the first stage,the level 1 decomposition with db4 mother wavelet is adopted to eliminate noise.Wavelet transform can separate low frequency information and high frequency information from stock price,reduce noise in data and optimize LSTM prediction results.In the second stage,Multi-input LSTM is adopted to predict stock price trend.The improved LSTM model can process multidimensional data and improve prediction ability by adding features and mining more information.In this paper,Chinese stock market data,US stock market data and technical indicators are used as inputs of Multi-input LSTM to forecast stock market.Then the model is compared with LSTM,decision tree,random forest,SVM and XGBoost models.The empirical results show that the accuracy of the proposed model is74.77%,which is better than other models.The combination of wavelet denoising and Multi-input LSTM can simplify learning process of LSTM,shorten training time and solve the problem of slow convergence of neural network model.In addition,in order to test practical application value of the model,this paper also selects representative FTSE 100,CAC 40,DAX,Nikkei 225 and NASDAQ to make predictions.It is found that the model has good prediction results,which reflects that it has universal applicability and broad application prospect in actual investment.Based on the above conclusions,this paper puts forward the following suggestions to investors,governments and financial regulators.First,encourage application of quantitative technologies to help investors and government departments make scientific decisions.For investors,the use of financial technology methods such as deep learning can help reduce risks.From the perspective of government departments,application of stock price prediction models can help determine stock price trends and economic conditions,prevent and resolve major financial risks,and maintain stability of financial market.Second,promote moderate regulation of financial sector and introduction of "sandbox regulation" mechanism so as to provide new regulatory ideas for safe and orderly development of China’s capital market.