Study On Option Pricing Based On Parametric Model And Nonparametric Machine Learning Model

The closed solution form given by the traditional parametric model brings great convenience to the solution of option pricing,but its pricing effect is not ideal because of its limitation in many aspects such as computing power and statistical hypothesis.Machine learning builds models by directly learning financial data of high-frequency trading and is not limited by other conditions.Therefore,it is reasonable to use the data mining ability in non-parametric machine learning models to improve the accuracy of option pricing.In this paper,we firstly estimate option prices by using a parametric option pricing model combined with GARCH volatility and historical volatility(60-day,90-day and 180-day standard deviations),and then compare the different estimation results.Secondly,four nonparametric machine learning model(Neural Network,Support Vector Regression,Random Forests and Boosting regression)used in option pricing,and introduces three methods of option pricing method and reference model(direct method,cascade method),the machine learning option pricing model is improved and the comparative analysis of different methods on the option pricing accuracy.Finally,the prediction results of the parametric model and the nonparametric machine learning model are compared,and the robustness of the results is tested.The main conclusions of this paper are as follows:(1)Parametric option pricing model can estimate option prices well to a certain extent.The prediction accuracy of historical 60-day volatility is the highest,followed by historical 90-day volatility and GARCH volatility,and historical 180-day volatility is the lowest.(2)The prediction accuracy of non-parametric machine learning model for option price is obviously better than that of parametric model.The prediction accuracy of Random Forest for option price is the highest,Boosting Regression and Support Vector Regression are second,and BP Neural Network has the largest error.The prediction accuracy of in-price option is generally better than that of out-of-price option.The larger the number of training sets,the smaller the error of the machine learning model,but the increase or decrease of the sample set has almost no effect on the parametric model.(3)The reference model method combines the advantages of traditional parametric method and nonparametric method,and the pricing effect is the best,followed by the direct method and the cascading method.The prediction accuracy of directly input underlying asset price and strike price is higher than that of value range.The input variable of the direct method using historical60-day volatility is the best predictor of option pricing.The innovations of this paper include: on the one hand,the predictive ability of parametric model and non-parametric machine learning model to option price is compared in detail.On the basis of neural network and support vector regression,random forest and Boosting regression are added to predict option price.On the other hand,an improved hybrid option pricing model is built through the reference model method,and how to improve the predictive accuracy of non-parametric machine learning model for option pricing is studied in a deeper level.