Bayesian Analysis Of High Frequency Financial Data Based On MCMC Algorithm

With the growing of electrical transaction in the stock market,high frequency data is easier and more convenient to attain.Moreover,the analysis of high frequency data has a significant effect on getting the dynamic information of micro-mechanism.So,the analysis of high frequency data is especially important.And this article is devoted to build price and duration model with the Bayesian method based on the characters of high frequency data.In this paper,Shanghai Pudong Development Bank(SPDB) stock data between March 21 and March 25 in 2016 is chose to analyze the characters of the high frequency data in China.Firstly,build an Bayesian ordered probability model for price changing according to the high frequency data on March 24.It can improve the speed of convergence by using the method suggested by Liu et al.(2000).After MCMC sampling,the parameters can be estimated by calculate the means and standard deviation statistics involved.Then utilize the estimated model to forecast the price changing,which finds that the model can catch 68%correct information.Secondly,build a threshold stochastic conditional duration model(T-GSCD) according to the data between March 21 to March 25 of SPDB.After adjusting the day mode,build a conditional duration models.And choose the best model G-SCD(1,2) through comparing the information criterion.Afterwards,with the testing of threshold method,it is found that G-SCD(1,2) model exists nonlinear property,which tells to build a threshold stochastic conditional duration model,which can explain the different dynamical correlation between large volume and small volume.Thirdly,build a two element model involving price and duration. Considering the SPDB stock data on March24,build a PCD model,and is estimated by MCMC algorithm.Themodel shows that price changing and duration exist significant dynamical correlations.Moreover,the longer the duration,the bigger the probability of the small trade volume and positive price change.Besides,With the difference in price changing direction,the number of change is also different.Promptly,the negative change will suffer more non-changing price information,and the positive changing will suffer more information about price changing.And the more the trade information,the stabler the price change.The innovative point of this paper can be honor to Bayesian method,which is utilized to estimate ordered probability model,T-SCD,PCD.At the same time,this paper is also devoted to extend the SCD model to T-SCD model by using the MCMC algorithm to estimate,which can capture the different state of duration.Moreover,this paper has also analyzed and explained the micro-mechanism for Chinese stock market.