Title: Estimating multiple quantiles via a reparametrized stochastic gradient algorithm
Authors: Tso-Jung Yen - Academia Sinica (Taiwan) [presenting]
Abstract: A method is proposed for simultaneously estimating several quantile regressions with the same covariates. The method reparametrizes predictors by allocating observed covariates to different containers according to whether they are positively valued or negatively valued. The resulting regressions will consist two sets of covariates that are positively-valued and negatively-valued, respectively. The reparametrization allows us to control the order of regression coefficients without considering the sign of the corresponding covariates. The method adopts an iterative scheme based on the subgradient algorithm for obtaining a solution to the estimation problem. The method is then applied to multiple quantile estimation for time series data.