Font Size: a A A

Quantile Regression Model And Its Application In Finance And Economy

Posted on:2017-02-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:N KangFull Text:PDF
GTID:1109330488993377Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
Financial and economic system is a complex system with typical complexity (such as non-linearity, non-stability, heterogeneity, etc.), while heterogeneity is affected by policy environment, market competition, technical innovation or other factors, leading to different rules and inherent qualities for the intersystem and intrasystem. Moreover, heterogeneity has become an important breakthrough for complex phenomena in financial and economic system. Compared with the classic mean regression, quantile regression (QR) provides a basic analysis tool in heterogeneous modeling which reveals the heterogeneity of explanatory variables on the response variable at different quantiles. Hence, quantile regression tends to provide better evaluation of the system operation pattern under various environment, which is of great value both in statistical theories and practical meaning.We selected "quantile regression theory model with application to financial economy" as the topic of this paper. Through integrating the discipline of management, statistics and econometrics and combining theoretical analysis, numerical simulation and empirical research, we detail the research as:first we aim to extend the classic quantile regression theoretically; then we carry out hot issue-focused research under the framework of quantile regression practically. The main work and innovations are as follows:1)For linear quantile regression modeling, we extend the error correction model from classic mean framework to the quantile of interest. First, we introduce quantile error correction model (QECM) and intend to propose model structure:parameter estimation, order identification, diagnostic test and model forecasting(including conditional quantile forecasting and density forecasting).Then, we manage to identify a better precision and accuracy of QECM model, utilizing the comparison of MECM model and quantile autoregressive model(QAR) via numerical simulations. Finally, we adopt QECM model to explore the relationship between money supply and price level, also we confirm the outperformance of QECM model in density forecasting.2) For non-linear quantile regression modeling, we carry out the analysis using the threshold quantile autoregressive model (TQAR). After proposing the model selection, we derive the asymptotic distribution of the threshold estimator and construct the likelihood ratio test of threshold effect. Moreover, through numerical simulation, we manage to spot that TQAR model outperforms in both threshold and parameters estimation, comparing to TAR, TAR-GARCH and QAR models. Also, a better precision and accuracy prediction has been obtained by the proposed novel model. Eventually, we adopt the proposed model with application to China stock market, in order to reveal the threshold effects and heterogeneity effects in the auto-correlation of return series. Meanwhile, we intend to apply TQAR model to assess the nonlinear dynamic characteristics of inflation and demonstrate the conditional density prediction of the fluctuation.3) For non-stationary time series quantile regression modeling, we aim to test the existence and heterogeneity of China Fisher effect based on quantile cointegrating regression. Specifically, we first estimate the real regression relationship between the nominal interest rate and inflation rate, and further propose the corresponding quantile error correction model to characterize the difference of adjustment speed of nominal interest rate from short-term fluctuation to the long-term equilibrium at different quantiles. There is evidence to show that more profound results can be obtained through quantile cointegration than traditional mean cointegration method. Moreover, we identify whether or not Chinese economic system holds Fisher Effect and indicate the level of it under specific states (quantiles). Also, we manage to describe the heterogeneity of Fisher Effect by measuring adjustment speed at different quantiles comprehensively.4)For "poor information" prediction modeling, combining the quantile regression with the grey prediction model, we propose the LAD-GM(1,1), LAD-GM(2,1), LAD-GM(1,1) power model and LAD-MGM(1, m) model and construct the parameter estimation based on the median regression. Compared with the traditional least square method, the median regression can not only increase the prediction accuracy effectively, but also overcome the deficiencies of the poor robustness and morbidity of the least square method, and improve the applicability of the grey prediction models.
Keywords/Search Tags:Quantile regression, Error correction model, Cointegration, Threshold autoregressive model, Grey prediction model, Conditional density forecasting, Heterogeneity
PDF Full Text Request
Related items