Recent experience has shown that interior-point methods using a log barrierapproach are far superior to classical simplex methods for computing solutions to large parametricquantile regression problems. In many large ...Recent experience has shown that interior-point methods using a log barrierapproach are far superior to classical simplex methods for computing solutions to large parametricquantile regression problems. In many large empirical applications, the design matrix has a verysparse structure. A typical example is the classical fixed-effect model for panel data where theparametric dimension of the model can be quite large, but the number of non-zero elements is quitesmall. Adopting recent developments in sparse linear algebra we introduce a modified version of theFrisch-Newton algorithm for quantile regression described in Portnoy and Koenker[28]. The newalgorithm substantially reduces the storage (memory) requirements and increases computational speed.The modified algorithm also facilitates the development of nonparametric quantile regressionmethods. The pseudo design matrices employed in nonparametric quantile regression smoothing areinherently sparse in both the fidelity and roughness penalty components. Exploiting the sparsestructure of these problems opens up a whole range of new possibilities for multivariate smoothingon large data sets via ANOVA-type decomposition and partial linear models.展开更多
基金This research was partially supported by NSF grant SES-02-40781.
文摘Recent experience has shown that interior-point methods using a log barrierapproach are far superior to classical simplex methods for computing solutions to large parametricquantile regression problems. In many large empirical applications, the design matrix has a verysparse structure. A typical example is the classical fixed-effect model for panel data where theparametric dimension of the model can be quite large, but the number of non-zero elements is quitesmall. Adopting recent developments in sparse linear algebra we introduce a modified version of theFrisch-Newton algorithm for quantile regression described in Portnoy and Koenker[28]. The newalgorithm substantially reduces the storage (memory) requirements and increases computational speed.The modified algorithm also facilitates the development of nonparametric quantile regressionmethods. The pseudo design matrices employed in nonparametric quantile regression smoothing areinherently sparse in both the fidelity and roughness penalty components. Exploiting the sparsestructure of these problems opens up a whole range of new possibilities for multivariate smoothingon large data sets via ANOVA-type decomposition and partial linear models.
