Research On Stock Price Trend Prediction Based On Deep Learning

In recent years,with the rise of big data technology,deep learning algorithm has become a more mature algorithm in the field of machine learning.At present,people generally choose the cyclic neural network(RNN)to deal with time series data first,because when dealing with time series data,RNN can obtain better prediction results with its unique short-term network memory advantage.But because of the problem of gradient vanishing and gradient exploding when RNN processes long sequence data,scholars pay more attention to long-term memory model(LSTM).LSTM,as an improved RNN model,its unique gate structure enables the model to process longer sequence data,and fundamentally avoids the problem of gradient vanishing and exploding in design.The empirical results of a large number of scholars at home and abroad show that LSTM deep neural network can simulate the cognitive process of neural organization more truly,has better processing ability for data with spatiotemporal correlation characteristics,is superior to the econometric analysis method,BP neural network model and ordinary RNN model,and is suitable for the research of stock price prediction.For the domestic stock market,as the domestic economic and trade environment is facing unprecedented changes and challenges,it is imperative to predict the stock market from a new perspective and method.Therefore,this paper applies new technology and new data set to the field of stock market trend prediction,hoping to design a more scientific and accurate stock market trend prediction model.Based on the improved LSTM model,this paper designs and models the up and down trend of the stock market,forecasts and verifies it by back testing.In order to improve the accuracy of the model,this paper compared with the previous scholars' research,in the input characteristics,training set selection aspects of the innovative design,considering three aspects of the comparative combination.In terms of input characteristics,this paper takes the extraction of market data,commodity prices and macro fundamentals as indicators,and takes their combination as the input variables of deep learning network for comparative analysis,and selects the appropriate combination of input characteristics;in terms of training set selection,through comparative analysis of the results of models under bull market and bear market cycles and forecasts based on different types of indexes Results,the characteristics of the model itself and the applicable scenarios are summarized;in terms of the network structure and parameter setting of the model,this paper selects the most suitable model structure through experiments with the accuracy as the index,and adopts the program adaptive modification setting for the important parameters of the model to avoid the subjectivization of parameter setting.In order to verify the actual effect of the model,we carried out quantitative back testing and found that the model performed well in the whole cycle and bear market cycle;in order to verify its effectiveness,this paper screened out a number of mainstream excellent quantitative funds in the market and compared their yields in the same period.The results show that the return of the model is much higher than that of index and quantitative fund in the whole cycle.At the same time,in order to improve the effectiveness of the application of the model in the bear market cycle and avoid negative return rate,this paper innovatively tests the three trading strategies combined with the model in a single capital cycle.The comparative results show that the three trading strategies improve the performance of the model.Among them,the reverse trading strategy has the least volatility,the largest revenue window and the most practical application space.Finally,the paper summarizes and explains the relevant characteristics and prediction results of the model,and puts forward the corresponding trading strategies and policy recommendations.