Research On The Application Of Split Parameter Estimation Methods

With the rapid development of time series econometrics,researchers have gradually found that when the time series is nonstationary,integration as traditional methods of processing the nonstationary of the results obtained in the test are not consistent,indicating that the current processing method of nonstationary still have development space,therefore,the researchers has been extended the integration to the fractional integration theory.Due to the presence of fractional integration theory and fractional integration model,when the time series exhibit long memory characteristics,reasonable use of the fractional integration method and appropriate ARFIMA model,memory characteristics can be further detailed description of the time series,in this process,the fractional integration parameter describing the time series as important parameters of memory,which is particularly important to the estimation of the fractional integration parameters.Although the existing literature has been widely used in various fractional integrated parameters of semiparameter estimation method to analyze the long memory time series of macroeconomic,stock market,,futures market and foreign exchange market,which is representative of the GPH,Sperio,LW and ELW,but did not point out the different types methods whether the model can get the best results in the semiparameter estimation of time series of,there are little compare between fractional integration model and integration model.This paper attempts to proceed from this perspective,through the research and application of modeling results of different estimation methods to analysis for macro economy in CPI series,Shanghai and Shenzhen 300 stock index series,which is the most suitable parameter estimation method for them,as compared with the integration application of modeling results.Firstly,the definition,the test method and the common model of the fractional integration are analyzed by theoretical analysis,and then describe the semiparameter estimation method.Secondly,we use real macroeconomic data and stock index data to analyze the changes of various methods under different bandwidth,Comparing the advantages and disadvantages of the four methods,we should avoid using GPH method and choose Sperio and ELW methods as much as possible.Finally,the D values estimated from the a=0.5 bandwidth are modeled by experience,and the results are compared in the fractional integration and the integration method model.According to the results of modeling analysis,we get the optimal parameter estimation method suitable for the series.The macroeconomic series is suitable for the Sperio estimation method,and the daily return of stock index is suitable for ELW method.The main conclusions obtained in this paper are as follows: 1.When choosing the actual parameter estimation method,we should avoid using the GPH method and select the Sperio and ELW methods according to the size of the sample data.2.when the parameter is estimated,the GPH method is avoided and the Sperio and ELW methods are selected according to the size of the sample.3.when the time series has long memory,the fitting effect of the traditional integer model does not have a good fitting effect of fractional integration.4.the CPI series in macroeconomics should consider data with long memory and analyze it by using the fractional integral method.Finally,it is concluded that the Sperio method is better than other methods to estimate the fractional integer parameters.5.the ELW method to estimate the diurnal yield series is better than the results obtained by other methods,and the result is obviously better than the integer model.