For a random variable Y with probability distribution function
the th quantile of Y is defined as the inverse function
For a random sample {y1, ..., yn} of Y , it is well known that the sample median is the minimizer of the sum
of absolute deviations
Likewise, the general th sample quantile ( ), which is the analogue of Q( ), may be formulated as the
solution of the optimization problem
where (z) = z( − I(z < 0)), 0 < < 1. Here I(·) denotes the indicator function.
Just as the sample mean, which minimizes the sum of squared residuals
can be extended to the linear conditional mean function E(Y |X = x) = x0 by solving
the linear conditional quantile function, Q( |X = x) = x0( ), can be estimated by solving
for any quantile 2 (0, 1). The quantity ˆ ( ) is called the th regression quantile. The case = 1/2,
which minimizes the sum of absolute residuals, corresponds to median regression, which is also known as
L1 regression