| A lot of strict assumptions are required for data themselves when we analyze Economic phenomena by modeling.Models and conclusion are trustworthy only these conditions are satisfied.For the data themselves,we often assume they are uniform and homogeneous,namely,we often assume that each point in data set are basically the same in effect for modeling,each point can affect the modeling,but their effect are small,a single point or several points should not be affect decisively for the general trend of the model.Actually the condition frequently can not be met.There are constantly one point or several restless points in a data set,these points often stir up trouble because there are flaw of the modeling means,they are outliers.what are influence of the outliers for capturing the market opportunities and modeling? How to identify and detect the outliers in data set? How to explain the cause and treat the outliers? What are the application of the diagnostic analysis of the outliers?The dissertation is divided into six chapters.The first chapter is introduction which include three parts,above all it give that the influence of outliers should be important consideration in data modeling,the next is literature review which give the progress of research.The final is the basic concepts of the outliers,research ideas and the main diagnostic statistics.The second Chapter details the effects of outliers for capturing market opportunities and several methods about the identification and detection of outliers.The third chapter analyzes the causes and treatment of outliers.The fourth chapter is empirical analysis which illustrate the application of the outliers diagnostics for the stock market.The fifth chapter is empirical analysis which illustrate the application of the outliers diagnostics for the corporate governance in the securities companies.The sixth chapter gives conclusion and prospect.The major work of the dissertation is as follows:1.The dissertation details the related theories about the outliers.In the paper there are several diagnostic statistics.There are a lot of outliers in actual data set.The current modeling methods are not perfect.GARCH model can not capture all the characteristics of the financial time series.There are serious and excessive skewness and kurtosis of the standardized residuals which are estimated by GARCH model.It is the outliers of the financial time series that lead to the phenomena.The outliers can cause the incorrect setting of the model,bad forecasts,bias of the parameter estimation and unreliable statistical inference.Therefore the identification and correction about the outliers should be significant consideration in the modeling of the financial data.There would be bring about loss of market opportunities or suboptimal market selection.2.The technology about identifying and detecting outliers is as follows: the LOF detecting method of outliers,the CUSUM detecting method of outliers,the wavelet technology about detecting outliers in financial time series.3.In the paper,we analysize deeply the cause of the outliers.The causes of outliers consists of systematic error,random error and gross error.There are some regular patterns in systematic error and the causes of its emergence would be known or grasped.Offsetting is the most essential characteristics of the random error,namely,as long as those error which contain offsetting can be regarded as random error.There is no impassable divide between the systematic error and the random error.Those error which had been classified as random error because of limited knowledge would be confirmed clearly as systematic error and process appropriately in technology.On the contrary,we may classify systematic error as random error and deal with the data through statistical method.4..In the paper,we investigate the treatment of the outliers.The removal of outliers should be rely on objective and reliable criteria rather than subjective judgement.Those standards which we may be able to adopt is 3s criteria,Chauvenet criteria and Grubbs criteria.5.Empirical analysis demonstrate that those outliers in data set have been located accurately,this method reduce our cost in investigation and trial.1).We selected three stocks which had been punished as samples,simultaneously,a regression model was established which the price was regarded as response variable and the earning ratio and the absolute value of the change was regarded as independent variable.We used data modeling diagnostic methods,according to the the diagnostic statistics of the studentized residuals,leverage values,cook distance,mahalanobis distance,then,we detected whether the price was abnormal and determined the suspect data by integrated crossing-confirms.After deleting the suspicious data,So we compared the characteristics about the qualities of model.Empirical studies have shown that the abnormal behaviors of the three cases had been located and conformed actual results,which was positive significance for the keeping the healthy securities market and protecting the legitimate rights and benefits of investors.2).In this paper,annual rate of return is regarded as dependent variable,average daily closing price etc.are regarded as independent variables,we established regression model by 46 main indices.We detected if data points are abnormal by Mahalanobis distance etc.diagnostic statistics according to data modeling diagnostic methods.Empirical studies have revealed that those indexes which were regarded as abnormal had been located accurately and we found three indices of China were not abnormal status,which were positive significance for understanding stock market in China.3).In this paper,earnings per share is regarded as dependent variable,per capita annual salary of the senior executives,per capita shareholding proportion of the senior executives and total share capital are regarded as independent variable,we established regression model by 58 companies which have complete annual report of Shenzhen Stock Exchange GEM in 2011,2010,2009.We detected if earning per share is abnormal by WK statistics,studentized residuals,Cook distance,leverage value and Mahalanobis distance according to data modeling diagnostic methods.We determined key suspect data by comprehensive cross affirmance and then we compared the changing of the several indicators before and after which represented the merits of the models after deleting these suspect data.Empirical studies have revealed that those companies which were regarded as abnormal had been located accurately and had been in line with actual results.All of these were positive significance for the keeping the healthy securities market,strengthening internal structure adjustment of the GEM listed companies and protecting the legitimate rights and benefits of investors.4).The Futures Companies are the main body of the futures market and their operating condition is directly related to the healthy development of the futures market.In this paper net profit is regarded as dependent variable,net capital and client's interests etc.are regarded as independent variables,we established regression model by 160 futures companies in 2011.We detected if data points are abnormal according to data modeling diagnostic methods.Empirical studies have revealed that those companies which were regarded as abnormal had been located accurately and had been in line with actual results.The conclusion is positive significance for the healthy development of the futures market and the corporate governance.5).In this paper,net profit is regarded as dependent variable,net income of acting as agent of securities business dealings etc.are regarded as independent variables,we established regression model by 94 securities companies.We detected if data points are abnormal according to data modeling diagnostic methods.Empirical studies have revealed that those companies which were regarded as abnormal had been located accurately and had been in line with actual results.The conclusion is positive significance for healthy development of the securities market and the corporate governance. |
PDF Full Text Request