Conditional Density Prediction Of Financial Market Via Support Vector Quantile Regression Approach

In the field of economic and financial quantitative analysis, accurately predicting the change law of economic and financial variables has significant value for important decisions for economic and financial policy and implementation of the corresponding control scheme. Due to the fact that economic and financial market are complex systems and their main performances are nonlinear, asymmetry and heterogeneity, traditional models and methods, such as traditional linear mean regression, are not enough to discover the running mechanism effectively.In this thesis, we proposed a new model called support vector quantile regression through combining support vector machine and quantile regression. We apply the SVQR model to financial quantitative analysis. The SVQR model can exert the strengths of support vector regression and quantile regression enough. On the one hand, it can reveal the nonlinear structure of the economic and financial system through support vector regression. On the other hand, it may describe the conditions distribution of the response variable by quantile regression, revealing the asymmetry and heterogeneity of economic and financial system. Therefore, the SVQR model can always get the desired empirical results in practice.In this thesis, the main contributions on support vector quantile regression theory and application are in the following there aspects. First, we apply the SVQR model to the money demand analysis and RMB exchange rate prediction respectively. We present a series modelling techniques including mathematical form, parameter estimation, the condition quantile prediction, and etc. Second, we propose the conditional density prediction method by differencing the conditional quantile. This method enables us to implement conditional density prediction of money demand and RMB exchange rate. Third, We compare the SVQR model with the traditional linear quantile regression through Monte Carlo simulation and empirical studies. The numerical results show that the proposed approach outperforms some widely used quantile regression methods in terms of the prediction accuracy. The numerical results also show that the SVQR model is not only able to discover the nonlinear relationship between money demand (or RMB exchange rate) and its influence factors perfectly, but also provide the conditional density prediction results which can predict the expected value of money demand and RMB exchange rate and determine the distribution and the shape that make it possible to consider how about explanatory variable influence the value of responsive variable. Therefore, this conditional density prediction method can provide more important information than traditional point prediction, which provides decision basis of policy about money demand and RMB exchange rate.To sum up, the research results are sufficient to show that the SVQR model not only is good at dealing with nonlinear system and describing the whole conditional distribution but also can reveal the complex relationship hidden in economic and financial systems. It has the potential to be applied in more and more new fields. In the future, the following two aspects are worthy to further study. First, in terms of the theory and method of modeling, we could consider constraints based SVQR through adding penalty terms to the SVQR model. The new SVQR model is able to handle the problem of many covariates through model selection procedure caused by the penalty terms. Second, we could discuss how to apply the information of conditional density prediction to decision optimization, such as portfolio selection decision,.