Font Size: a A A

Deep Learning Of Barter Trade Index:Economic Pricing,Statistic Identifying And Data Driven

Posted on:2018-12-04Degree:DoctorType:Dissertation
Country:ChinaCandidate:J H HeFull Text:PDF
GTID:1319330518964803Subject:Statistics
Abstract/Summary:PDF Full Text Request
This paper focuses in Textele Price Index including Total Price Index,which covering Row Materials,Grey Cloth,Apparel Fabric,Home Textile and Fashion Accessories,that's five broad headings.And each broad heading covers several medium headings.Textel Price Index holds the balance in the index system,since the price the focus for the industry and the community.So it is significant to do further quantitive analysis for the development of Keqiao Textile Index and the textile industry.A sound and consummate "CHINA·KEQIAO Textile Index" is infavor of the government to know and supervise the development trend of the industry,and thus to provide the basis to make right decision.It is advantageous to the enterprise of textile production and market merchants to provide information services.Given the circumstances,oriented by the technic this paper utilizes the technical indicators to transit description statistics to inference statistics,to transit classical situational analysis to verifiable model forecasting,based on the concepts of big data,from economical pricing to traditional statistic identification,to data learing till index reconsitution.The study begins with economic pricing,then statistic identification,which is from circumscribed economic pricing to rigorous statictics hypotheses.Then,it is the looser data-driven pattern.Finally,data learning and true time forecasting are done.The paper studies Textile Price Index with multidisciplinary methods;they are econometrics,statistics and mathematics.The properties of the Textile Price Index population are deduced by statistics tests.The factors that influence the Textile Price Index signefecantly are selected and identified by nonparametric regression completely data driven.It prices the Textile Price Index by classical CAPM,then statisticall price it by nieve model and time series model.Textile Price Index is forecasted by state space model and Kalman filtering.Contrastive analysis and descriptiveness are throughout the empirical analysis in the article.The Testile Price Index is simulated by establishing different models.The simulating results are demonstrated by descriptiveness.The inpacts to the Index from various factors,the different predictive features of the models are explained by contrastive analysis.The parameters of the population model are estimated by exploratory organon.The direct and indirect inpacts to the explained variable made by the nonlinear variables are revealed by nonparametric path analysis.All the caculations are realied by software EViews 6 and statistical software R codes.The research contents are as follows.(1)To orgnize the models prcing financial assets and real assets,especially the application of CAPM.(2)To prices the Textile Price Index by classical CAPM in the sense of economics.Since there is no risk-free return as the model hypothesis,the index is repriced by zero-beta CAPM.(3)The sample from the time series of Textile Price Index is analyzed and tested by statistical methods,including descriptive statistical analysis,normality test,ADF and KPSS tests,autocorrelation test,run test,independent indentically distribution test.All the mathematical models,as well as statistical models,have certain conditions.Before modeling practical issue,it is necessary to test whether the conditions are true.If the conditions are not true,the model is un reliable and the results deduced from the model are worthless.(4)To model the Index with data driven method.There should be no practical issues in conformity with the model conditions completely.Usually,the conditions are approximately satisfied,the more approximage the more reliable the results are.The paper adopts complete data driven nonparametric methos to select signefecant factors from the set consiting of the factors possibly influence the Index,and identify the agent variables,linear or nonlinear.Later,establish models to analyze each of the factors's strength infecting the Index.(5)To construct model for forecasting.Within a reasonable model system,there would be forecasting function.CAPM has rigorous hypothesis difficult to be tested.Before forecasting with semi-parametric models,have to predict value of the linear and nonlinear independent variables.So the models' forcasting function is theoretic.To resolve the problem,construct state space model to realize real-time prediction to the Textile Price Index with Kalman filtering.After all the works,obtain the results:(1)The results from economic pricing.We apply CAPM to pricing index of real asset trading.After modeling the weekly index and monthly index,we find the weekly model is better than monthly ones.In the sense of statistics,the determination coefficients of the monthly equations are bigger than weekly equations'.Economically,the intercept term can not reject the null hypothesis that equal to zero,meaning that the price is reasonable.The coefficient of the first degree term is not zero signefecantly,meaning that the price is influenced by market risk.But there are 9 of the 30 medium categoris' models make sense.The results from the zero-beta CAPM are similar.The reason is that the market conditions and the traders' expections are far cry from the hypotheses of CAPM.So it is necessary to uncover the properties of the Index population by sampling.(2)The results from statistic identification.We test the population with the teim series sample which period is from May,2007 to December,2016.ADF test and KPSS test show that the series is a first-order unit root process.Auto-correlation test shows the processe is autocorrelated.Run test uncovers that it is not a random walk processe.Of course,BDS test finds it is not independent indentically distributed.The conclusion is that the Textile Price Index series is not fit for models with rigorous hypotheses.(3)The results from data driven variables selecting and identifying.Based on selected variables such as monthly average crude oil price,consumer price index,retail price index,producer price index and benchmark one-year deposit rate are introduced as relemant factors with Textile Price Index by nonparametricpath design,and it is identified that consumer price index and producer price index influence the dependent in linear pattern still by nonparametric approach.Because of the lagged effect of the Tetile Price Index the lagged is employed as linear independent variables.Therefore we construct a path desing model with four linear independent variables and four control variable,namely non-linear independent variables.(4)The results from semiparametric model.We establish varying-coefficient partially linear models with all the significant variables as dependent variables and the Textile Price Idex as the independent variable,and find that each of the parametric terms accounts for 20%,while the nonparametric term is slingtly higher than 20%.Cotton index A plays the most important role of the sic control variables.The empirical and simulated results reveal that none of the ratios would be extreme as long as cotton index A does not exceed twice the sample mean during thd sample period.(5)Establish real-time forecasting model.As a complete model system,it should be able to forecast.Using three kind of model,they are naive model,time-series model and state space model(Kalman filtering),we fit the Textile Price Index observations and get the latest index value prediction.During the sample period,the error is measured by error of mean squre root.We find the all the error is acceptable.The time-series model is better than the naive model,which illustrates the time series model is correct.The early recurrence prediction of the state space model is related to the initial value.The influence will eliminate after enough recurrence,so after abandoning the early 60 forecast,its accuracy is better than time series model's.Whatever the prediction accuracy or computability,state space model is the best of the three.
Keywords/Search Tags:CAPM model, economic pricing, data-driven, statistic identification, state space model
PDF Full Text Request
Related items