CEV Option Pricing Model Based On Neural Network And GARCH Model

Modern option pricing theoretical research is mainly divided into two types:parameter models and non-parameter models.Most existing parameter model establishment mainly relies on many market assumptions.Related assumptions often have an excessive state,which makes the results and real situation errors.At the same time,due to its rigorous logic,it is limited to the connection between variables,and it is easy to form inherent thinking,and there are limitations.The infrastructure is limited by self-cognition and cannot fully reflect the real situation of the market,causing errors in the prediction results.Nowadays,the emergence of non-parameter option pricing models such as neural networks completely depends on the characteristics of the true trend of the market,which has triggered a new attempt to make many scholars.Therefore,in order to discover the advantages of the pricing results of traditional pricing formulas and neural network models,it is necessary to conduct empirical analysis of different pricing models.At the same time,because the neural network model has pure data driving and good adaptability,this needs to control the accuracy of the input data and the proper control of the neural network learning process to achieve complementary traditional models and general neural networks.First,select and preprocess the pricing data of Shanghai Stock Exchange 50index options;Then carry out sample analysis and correlation test,and establish the optimal GARCH model after confirming that the correlation test has passed;Second,through the estimated volatility,the parameters in the CEV modelbecome time series;Thirdly,by comparing the CEV model under fixed parameterand random parameterβ,this thesis proves that the introduction of time seriesand the corresponding parameters of each moment in the CEV model can optimize the CEV model_（80）.Fourth,based on the control variable method,this thesis compares the pricing effect of the LSTM model and the CEV model under the sameβ,which proves that the option pricing effect under the existing parameters and information can be optimized considering the LSTM,and the pricing effect of the LSTM model under the comparison of fixed parameterand random parameterproves that the pricing effect of the LSTM model under theof random parameters is better.Finally,the CEV model and the recurrent neural network model are combined to construct a hybrid neural network,and by comparing the pricing effect of the original neural network model and the hybrid neural network,it is proved that under the constraints of the traditional CEV model,the random performance of the neural network model is controlled to a certain extent,and the phenomenon of overfitting the neural network is corrected to a certain extent,so that the option pricing results are further improved.