The Risk Prediction Of Stock Market Based On Deep Learning Models

In today's society,science and technology are changing with every day,and artificial intelligence is developing in full swing in various fields.Machine learning models and deep learning models have been successfully applied on face recognition,speech recognition,natural language processing and other fields.At the same time,with the continuous innovation of financial technology,deep learning methods are used in the stock market more and more widely.This article analyzes and discusses the prediction of stock market risks based on deep learning models.The reasons for the formation of stock market risks are changeable and there are many influencing factors.We select 15 variables to build a model from three dimensions of stock market transaction fundamentals data,statistical technical indicators and extended interval data to predict stock returns.The volatility of stock returns is a signal of stock market risk.Therefore,the model for predicting stock returns can be improved to achieve a better prying into the practical significance of market risks.This article uses the relevant data of the Shanghai and Shenzhen 300Index from 1^th October,2008 to 30^th September,2020,by python language,Tensor Flow,and Keras frameworks,with the model of moving average,MLP,and LSTM to predict stock returns.In the moving average model forecasting,there are one-day sliding model by using the current day's value as the forecast value and N-day sliding average forecasts,and then the optimal N value is selected to construct the forecast model.Through the comparison of the four prediction results,the four models are gradually increasing in complexity,and the prediction effect of the model is getting better gradually.The root mean square error of the loss function obtained by the four methods is 0.02197,0.01610,0.01451,0.01380 respectively.The model effect increased by 37.2%.Subsequently,the hyperparameter training optimization is performed based on the LSTM model.The optimized hyperparameter variables include the time window length N,the number of neurons mlp units,the probability of dropout,the optimizer,the training period epochs,the number of small samples used for each parameter updating and function activation.Then the optimization model is constructed with each optimal hyperparameter under the selected range.By parameter optimizating,the model effect increased by 3.04%.According to the same optimized process,continue to predict the standard deviation of the rate of return,then obtain the conditional probability distribution of the rate of return to measure the value at risk Va R.Select the sample quantile as the early warning line,and establish an optimized risk early warning model.When the value at risk Va R is below the warning line,a risk warning signal is issued.This paper finds that in the offline period of the warning line,the true rate of return is continuously low,and the model's risk warning has achieved good results.The stock market risk forecast is based on a good forecasting model.When snooping on the signal of the risk,take active defensive measures.For individuals,try to reduce losses as much as possible.For financial institutions,stabilize the market.For society,reduce the impact on people's livelihood..