spacer picture
spacer picture

Quantile Regression

spacer picture

What is Quantile Regression?

Quantile regression is a statistical technique intended to estimate, and conduct inference about, conditional quantile functions. Just as classical linear regression methods based on minimizing sums of squared residuals enable one to estimate models for conditional mean functions, quantile regression methods offer a mechanism for estimating models for the conditional median function, and the full range of other conditional quantile functions. By supplementing the estimation of conditional mean functions with techniques for estimating an entire family of conditional quantile functions, quantile regression is capable of providing a more complete statistical analysis of the stochastic relationships among random variables.

Source: XKCD 2048 as amended by Anton Antonov for a 2019 talk at an R-user meeting in Boston.

Some References

Three elementary introductions to quantile regression:

Koenker, R. and K. Hallock, (2001) Quantile Regression, Journal of Economic Perspectives, 15, 143-156.

Cade, B. and B. Noon, (2003) A Gentle Introduction to Quantile Regression for Ecologists. Frontiers in Ecology and the Environment, 1, 412-420.

Hao, Lingxin, and Daniel Q. Naiman, (2007) Quantile Regression, Sage Publications.

A more extended treatment of the subject is also available:

Koenker, R. (2005) Quantile Regression, Econometric Society Monograph Series, Cambridge University Press. Errata list.

A even more extended treatment of the subject is now also available:

Koenker, R., Victor Chernozhukov, Xuming He, Limin Peng (eds) Handbook of Quantile Regression.

Applications


There are a wide variety of applications among which I've collected a few examples:





Software


Quantile regression software is now available in most modern statistical languages.
The recommended statistical language for quantile regression applications is R. R is a open source software project built on foundations of the S language of John Chambers. Capabilities for quantile regression are provided by the "quantreg" package. Once R is installed on a networked machine packages can be easily installed using the command install.packages("quantreg") in an R session. The documention for the quantreg package for R is available in pdf format from the CRAN website. A vignette describing the functionality of the quantreg package is also available. Some frequently asked questions about the quantreg package are answered in the FAQ. Some slides on the early history of quantile regression are available here. Some slides for a tutorial talk about QR computation are available here. A development package for quantile regression methods for longitudinal data is available from R-forge, it is called rqpd.

Various historical versions of the R package "quantreg" are available here. Some of these versions are available from CRAN, the primary R repository for packages. Some intermediate versions not available in the CRAN archive were removed from CRAN due to uncertainties about the licensing status of the fortran code for sparse Cholesky decomposition in cholesky.f. They are made available here for the sake of completeness and in the interest of reproducibility. I am pleased to report that these licensing uncertainties have now been resolved as the authors of cholesky.f, Esmond Ng and Barry Peyton, have declared their code to be open source, so subsequent versions of the quantreg code > 4.77 are freely available under the terms of the GPL license. See the README file in the package for further details.

A basic version of the interior point (Frisch-Newton) algorithm for quantile regression developed for the R quantreg package is also available for matlab. A version of the same algorithm that allows linear equality constraints is available here. A C++ translation of the algorithm is also available from Ron Gallant's libcpp library. This algorithm is described in Koenker and Portnoy (Statistical Science, 1997).

The original simplex code for quantile regression was based on the Barrodale and Roberts (1974) code, as described in Koenker and d'Orey (1987, 1994). Unfortunately the fortran code for these implementations which is available from the CRAN quantreg package is quite difficult to read and unravel, so at some point I wrote a mimimalist version of this algorithm in pure R that I hope will be useful to anyone seeking a more straightforward description. This version is available here.

Quantile regression is also available in the standard distributions of Stata, Shazam, Limdep and a number of other statistical/econometric packages. The Xplore language includes quantile regression functions based on the R code mentioned above. SAS now also includes a quantreg procedure modeled closely on my R quantreg package.

spacer picture

Econometrics
University of Illinois at Urbana-Champaign
Commerce West, Champaign IL 61820
contact Roger Koenker (roger@ysidro.econ.uiuc.edu)

spacer picture