Research On Financial Riskmeasurement Based On Quantile Autoregression

Financial risk increases with rapid development in financial markets. In recent years, financial risks occur occasionally, such as the global financial crisis from 2007 to 2009, the full-blown European debt crisis in 2010. People gradually realized that the internal uncertain factors are increasing, as well as the complexity of the market (such as nonlinearity and heterogeneous, etc.) since these financial risks. Time series analysis plays an important role in financial risk measurement. Based on the latest domestic and foreign research in this field, this thesis focuses on solving nonlinear and heterogeneous and other complex issues of financial system, selects the subject of "research on financial risk measurement based on quantile autoregression", extends time series analysis to the quantile regression framework, investigates nonlinear quantile autoregression of univariate time series, as well as quantile vector autoregression of multivariate time series, and then proposes the corresponding financial risk measurement methods.This thesis adopts mathematical analysis, artificial intelligence, numerical simulation and empirical research methods to carry out research from the following two aspects: First, in terms of theoretical modeling, extend the classical quantile regression models, and propose new financial risk measurement tools and methods; Second, in terms of research in practice, improve the efficiency of financial risk management by ustilizing new proposed methods. The characteristics of this study are embodied in two aspects. First,carrying out research in the quantile regression framework can reveal the full distribution of response variables and reveal the heterogeneity effect, improve the efficiency of tail risk characterization. Second, through the method of artificial intelligence such as artificial neural network, this thesis proposes the models and methods of financial risk measurement based on quantile autoregression, which can fully simulate the nonlinear characteristic of financial system, improve the robustness of financial risk measurement and realize the accurate measurement of financial risk. The main innovations of this thesis are as follows:1. Propose a quantile autoregressive neural network (QARNN) model and a new VaR risk measurement method. Firstly, the QARNN model is established by introducing the neural network structure into the quantile autoregressive model, with parameter estimation methods studied. Secondly, the AIC and GACV criterion of QARNN model order are established to determine the optimal lag order of the model; Thirdly, through the numerical simulation, the fitting effect and predictive ability of QARNN model were studied and compared with the traditional VaR methods, such as the RiskMetrics model, ARMA-GARCH model,CAViaR model,CARE model and quantile regression methods. Fourthly, apply the QARNN model to the financial risk measurement, propose the new VaR risk measurement method, and obtain a more satisfying financial risk measure effect.2. Propose a nonparametric conditional autoregressive Expectile (NCARE) model and a new ES risk measurement method. Firstly, establish the NCARE model by introducing the neural network structure to the Expectile autoregressive model, with parameter estimation methods studied. Theoretically prove that NCARE can get the consistent results of Expectile estimation. Secondly, the GACV criterion of NCARE model order is established to determine the optimal lag order of the model. Thirdly, the NCARE model's fitting effect and forecasting ability are studied by numerical simulation,and compared with the traditional methods. It is found that the former can better describe the nonlinear dynamic features of various financial time series. Fourthly,the NCARE model is applied to financial risk measurement, and the new ES risk measurement method is proposed, with emphasis on its distinguishing performance before, during and after the financial crisis.3. Quantile vector autoregressive distributed lag (QVARDL) model and its impulse response analysis are established to explore financial risk infection. Firstly, extend the vector autoregressive distributed lag model to the quantile system, propose QVARDL model, and a set of modeling methods such as mathematical representation, parameter estimation, model order and impulse response analysis are presented. Secondly, the conditional quantiles of multiple time series at various quantiles are estimated and the dynamic relationship between them is given; propose the measurement method of financial risk infection based on the quantile impulse response analysis. Thirdly,empirical studies are conducted on the capital markets in major countries (regions)through the proposed QVARDL model to investigate the impact of American subprime mortgage crisis. The empirical studies are conducted mainly through the multivariate and autoregressive approaches, and grasp the law of variation of the conditional quantile from both horizontal and vertical aspects. It is helpful to comprehend the spreading of subprime mortgage crisis in America from the aspects of influence degree, influence mode and response time.Overall, firstly the classical quantile regression models are expanded to develop new quantile regression models and methods, which enrich the research content of quantile regression theory. Secondly, the thesis chooses the hot issues in the field of financial risk measurement, carries out the related subject research in the framework of quantile regression, and focuses on the study of the heterogeneous and nonlinearity characteristics of the financial market, and solves the problems of accurate measurement of financial risks. The research not only helps financial institutions enhance risk management to improve international competitiveness, but also helps the supervision department supervise in order to ensure the healthy development of the financial market.