| Financial volatility modeling has always been a hot spot and focus in the field of financial risk research.With the rise of Internet finance,its volatility has greater uncertainty than traditional finance.Therefore,it is more necessary and important to establish a variety of financial volatility models,and to evaluate and select the models.This article first puts forward the evaluation method of financial volatility model from multiple angles;secondly,through the empirical research on the volatility of Pacific stocks,constructs various Bayesian financial volatility models;finally applies the model evaluation method in this article to evaluate and apply it,And verify the rationality of the evaluation method proposed in this paper.The main work of this paper is:The first part is the evaluation method of financial volatility model.This article evaluates different financial volatility models from three perspectives.First,in terms of the model’s ability to characterize typical features,not only the kurtosis and autocorrelation function of the model are described separately,but also an evaluation method based on random simulation and the number of tail bits to characterize the model’s tail features is proposed;In terms of model fitting ability,considering that likelihood ratio hypothesis testing and Bayes factor calculations have certain difficulties,the concept of model fitting rate based on the ratio of the number of fitted samples to the total number of samples is proposed.The calculation of depends on the stable sample path obtained by the model randomly simulated enough times.The calculation process is simpler and the maximum fitting error of the model can be adjusted according to the boundary function range of the sample path.Finally,in terms of model prediction ability,Taking into account the different actual needs of model users and the asymmetric influence of the positive and negative fluctuation forecast errors,a weighted evaluation loss function L((?)_t,λ)based on the size of theλfactor to improve the model’s oversmoothing and undersmoothing is constructed.When multiple models may reach inconsistent conclusions under different evaluation angles,this paper integrates the index values of the previous three evaluation angles to establish a new model evaluation selection index,based on which the optimal model can be selected.The evaluation method proposed in this paper is a supplement to the existing evaluation methods and provides a theoretical method basis for model evaluation and selection for the following empirical research.The second part is the Bayesian volatility modeling part in the empirical research.Taking Taiping stock as an example,based on the analysis of the basic statistical characteristics of the sample data and the characteristics of the model,four types of volatility models,GARCH and SV,with error terms subject to different distributions are considered.In order to ensure the consistency of the parameter estimation methods of multiple model evaluations and the accuracy of parameter estimation,the Bayesian method is used to estimate the models,that is,the prior distribution of the model likelihood function and the parameters is first determined,and then derived The joint posterior distribution of the parameters and the full-condition posterior distribution of each parameter.Finally,the Gibbs algorithm is used to perform the posterior inference of the parameters,so as to complete the construction of the four Bayesian financial volatility models of Pacific stocks,which is the third part of the evaluation application Provide model preparation.The third part is the appraisal application part in empirical research.The four established Bayesian volatility models are evaluated and applied using the evaluation method proposed in this article,and they are found to have strong evaluation capabilities,and the results are verified by the kupiec test.Finally,It’s found that the Bayes-SV-T of Pacific Stocks is a relatively optimal model.According to this model,it is found that during the sample period,the volatility of Pacific stocks has changed greatly.The fluctuation amplitude in the last year is obvious,and the difference between the two largest fluctuations is 335 days;for the volatility forecast for the next100 days,it will remain On the fluctuation level of 0.02±0.01. |
