36
28

Monotone Least Squares and Isotonic Quantiles

Abstract

We consider bivariate observations (X1,Y1),,(Xn,Yn)(X_1,Y_1), \ldots, (X_n,Y_n) such that, conditional on the XiX_i, the YiY_i are independent random variables with distribution functions FXiF_{X_i}, where (Fx)x(F_x)_x is an unknown family of distribution functions. Under the sole assumption that xFxx \mapsto F_x is isotonic with respect to stochastic order, one can estimate (Fx)x(F_x)_x in two ways: (i) For any fixed yy one estimates the antitonic function xFx(y)x \mapsto F_x(y) via nonparametric monotone least squares, replacing the responses YiY_i with the indicators 1[Yiy]1_{[Y_i \le y]}. (ii) For any fixed β(0,1)\beta \in (0,1) one estimates the isotonic quantile function xFx1(β)x \mapsto F_x^{-1}(\beta) via a nonparametric version of regression quantiles. We show that these two approaches are closely related, with (i) being more flexible than (ii). Then, under mild regularity conditions, we establish rates of convergence for the resulting estimators F^x(y)\hat{F}_x(y) and F^x1(β)\hat{F}_x^{-1}(\beta), uniformly over (x,y)(x,y) and (x,β)(x,\beta) in certain rectangles as well as uniformly in yy or β\beta for a fixed xx.

View on arXiv
Comments on this paper