The Bayesian Expectile Regression And Its Empirical Research,

The traditional least square estimation only gives the mean of the dependent variable, but quantile regression and expectile regression can estimate the whole distribution of it. Based on expectile, a derivative of quantile and bayesian statistics, this paper presents the bayesian expectile regression based on Metropolis-Hastings sampling method. It shows the practicability of non-information prior in this regression. Two simulated data sets were used to evaluate the proposed method. The first one showed a good fit and the second one showed the similarity and difference between the expectile and the quantile. Finally, based on CAPM, employing data from Shanghai composite index in 2010, this paper displayed the advantage of expectile regression against the traditional least square estimation, and proved its feasibility and insufficiency.