Efficient estimation in expectile regression using envelope models (Q2286363)

The authors study expectile regression as an alternative to quantile regression. For \(\pi \in \left(0,1\right)\), the \(\pi\)th expectile of a random variable \(Y\) is given by \[ f_\pi \left(Y\right)= \arg\min_{f \in \mathbb R} \mathbb E \left[\left|\pi - I\left(Y-f < 0\right)\right| \left(Y-f\right)^2 \right]. \] For \(\pi = 0.5\), this equals the classical expectation \(\mathbb E\left[Y\right]\). This paper treats a new method for estimation of parameters in expectile regression, the so-called envelope expectile regression, which can be performed using the generalized method of moments. The authors establish consistency of the estimator and derive its asymptotic distribution. In addition, it is shown that the new estimator is asymptotically more efficient than classical expectile estimators. These results are illustrated by numerical experiments and real data examples.