Abstarct: Quantile regression a complete descxription of conditional dependence of Y on the explanatory variables. In ordinary regression model the dependence of conditional mean of response variable Y on explanatory variable X is checked. Parameters estimation in quantile regression is baxsed on an asymmetric loss function and calculates in the same way of least square methods. The idea of Bayesian quantile regression employing a likelihood function that is baxsed on the asymmetric laplace distribution. Binary and tobit quantile regression can both be viewed as linear quantile regression with a latent continuous response which is incompletely observed. Normally the time series of error is considered normal. In this thesis the assumption of non gausian of the error and have used it as a semi-parametric exponential function category considered when we set abnormal changes proposed and the simulation data sets describe building and two real series.