Research On Quantitative Stock Selection Model Based On Neural Network

From February 12 to March 16,2020,the US Dow Jones Index fell from the highest 29568.57 to 18917.46 in 35 days,a drop of 36.02%.During this time,there have been many meltdowns,and many investors faced huge financial problems,even bankruptcy.The difference is that the medal fund founded by Simmons,in March of the mad fuse,the medal fund received a 9.9% pure profit return,and the fund's secret to avoid losses is actually quantitative investment.As one of the branches of quantitative investment,the use of neural networks to predict stock price trends has become a hot spot and focus in the past few years.This article mainly studies the application of neural network in quantitative stock selection.For all A shares for the four years from February 1,2016 to February 1,2020,the three stocks that performed better were selected by predicting the trend of A shares in the next week.In order to meet the forecast of stock price trends within a week,this article selects five technical indicators(highest price,lowest price,opening price,closing price,and turnover rate)that affect stock price trends.By using the data of the past 34 days and the past 30 weeks The data turns the one-dimensional time series into a 64×5 two-dimensional data,then extends forward in steps of weeks,and generates the training set and the verification set by windowing.This paper studies the influence of parameters such as the number of hidden layers,batches,iterations,and optimizer on the prediction results of the neural network model and optimizes the corresponding parameters of the model,and uses these two neural networks to predict the rise and fall of individual stocks.It shows that both neural network models have good effect on stock price prediction.Finally,an A-share stock selection strategy is constructed.Based on the stock price prediction model in the previous chapter,a quantitative stock selection model based on neural network is further formed.And compare the two neural network stock selection models with technical index models.In terms of strategic earnings,cumulative returns,and outperforms on the Shanghai Stock Index and other earnings data,both neural network stock picking models perform more than three times better than traditional technical indicator stock picking models;and risk indicators such as Sharpe ratio and maximum drawdown rate Above,both neural network stock selection models are more than 70% better than traditional technical index stock selection models.In summary,the various returns of the two neural network stock selection models perform very well.The high Sharpe ratio and the low maximum drawdown rate both indicate that the risk is also very low,which indicates that the quantitative stock selection model based on the neural network It is effective and can bring excess benefits.