Application Research Of Deep Learning In Hedging In Stock Market

Since the father of deep learning in 2006,Jeffrey Sinton and his student Ruslan Salakhutdinov formally proposed the concept of "deep learning",deep neural network(DNN)technology is global.Deep neural networks are the basis of deep learning,and its essence is a neural network with multiple hidden layers.Deep neural networks are the basis of many artificial intelligence applications.Due to the breakthrough development of deep neural networks in data analysis,speech recognition and image recognition,the use of deep neural networks has exploded.The deep neural network obtains specific features by processing the original data,forming abstract features by combining specific features,and then forming more high-level abstract features by abstract feature combinations to represent the attribute categories or features of the learned objects,and finally establishing on the abstract features.Start the classifier and get the predicted output of the model.These deep neural network models are deployed in a variety of applications ranging from autonomous vehicles to cancer detection to complex games.In recent years,deep neural network models have also been applied to financial markets.Therefore,this paper attempts to use the deep neural network model to improve the prediction accuracy of the least squares model(OLS)for hedging ratio.Looking back at the previous researches on hedging by domestic and foreign scholars,most of them choose to use stock index as the spot and hedge the spot with the corresponding stock index futures.The conclusion is: Hedging the corresponding stock index with stock index futures.Preservation will have a good hedging effect.However,from the actual situation,since the stock index is not an index that can be traded,and because most individual investors have less capital and cannot fully construct the stock basket according to the stock index.The stock index used as the spot for hedging for most individual investors are unrealistic.In this paper,the author tries to use the food and beverage constituents in the fund's Awkwardness stock as an example,and hedges it as a spot,and improves the prediction result of the least squares model through the deep neural network model,bringing investors a new model for predicting hedging ratios.In the empirical study,this paper uses six deep neural network models with different parameters to predict the residual of the predicted hedge ratio and the actual hedge ratio calculated.After obtaining the hedging ratio residual,this paper divides the residual samples into training sets and test sets.The deep neural network model with six different parameters makes the model search for the sample law,determines the model parameters,and then predicts the test set samples with the learned model.Finally,records and organizes the six model training sets and test sets.The relative improvement rate of the results based on the mean square error(MSE)index of the least squares model and the relative improvement rate of the MSE indicators based on the six models.The empirical results show that the deep neural network model can improve the prediction accuracy of the least squares model for the hedging ratio,and the activation function Sigmoid function in the deep neural network is superior to the Re Lu function in future prediction;the deep nerve processed by the selfencoder is more predictive than the deep neural network model that has not been processed by the self-encoder.