2014年上海市扬帆计划见面会.ppt

Empirical application The data covers a period from January 4, 2002 to December 30,2016 are take from Wind Database. * Mathematically, this was formulated using linear regressive models of X and Y. In his framework, if the prediction of Y can be improved by incorporating the history of X, it is said that there exists a causal connection from the time series X to time series Y. * As we know, the mean regression models would result in better estimates of the effect of time series on each other, but it has many limitations in econometric applications, such as it may fail to detect causal relation if the distribution of the variables involved fat tailed or nonlinear causal relationships. * The quantile regression may be a powerful alternative. In recently, a useful and flexible modeling strategy that the Granger causality based on parametric quantile models has been proposed at all quantile levels by Troster in 2016. * A shortcoming of the standard framework for pairwise Granger causality is that it is difficult to identify whether the causal influence of one time series Y on another X is directed or mediated by another recorded time series. Although GCQ can get a relatively good causal relation under pair time series, it failing to identify causal patterns for multivariate time series. * Therefore, for three or more simultaneous time series, the causal relations between any two of the time series may be direct or indirect by a third one. Our proposed approach, conditional Granger causality in quantiles maybe a good choice to address this situation. * Hong Cheng Shanghai Lixin University of Accounting and Finance 2018.05.19 School of Statistics and Mathematics hcheng@cims.nyu.edu Conditional Granger-causality Test in Quantiles 2018 Symposium on Data Science Statistics Collaboration with: Tinggan Yang, Yuhua Luo Causality Clive W. J. Granger Born 1934 in Swansea, Wales (British citizen); Ph.D. from University of Nottingham in 1959; Emeritus Professor of Economics at U