The Volatility Model Which Is Based On Deep Neural Network

With China's financial market becoming more and more prosperous and the increasing per capita disposable income,various financial products are playing an increasingly important role in People's Daily life.The risk assessment of financial products has attracted wide attention,predicting risks has become one of the important topics for market participants.In the financial field,volatility is often used to reflect the risk value of the market.The traditional volatility model has a simple structure and a limited fitting ability,which cannot meet the needs of the increasingly complex financial market.In recent years,the deep neural network model is favored by many researchers because of its complex structure and strong fitting ability.Deep neural network has made remarkable achievements in many tasks and greatly promoted the development of artificial intelligence.In this paper,we combine GARCH model and deep neural network(DNN)to model volatility.We replace the volatility equation of GARCH model with a deep neural network,determine the loss function of the model by the maximum likelihood method,and update the model parameters by using the adaptive momentum estimation(Adam)method which is famous method in deep learning field.In addition,the model in this paper adopts the deep neural network with various structures such as fully connected neural network and circulatory neural network,and we use 5 fold cross validation to select the optimal model structure.This paper conducts an empirical study on daily return data of Shanghai stock exchange index from 2005 to 2019,and shows the comparison between the deep neural network model based on different structures and the traditional GARCH model.The results show that the volatility model which is based on the deep neural network is an effective improvement of the traditional volatility model and has better performance in predicting the volatility accuracy.