35
13

Regression estimation from an individual stable sequence

Abstract

We consider univariate regression estimation from an individual (non-random) sequence (x1,y1),(x2,y2),...×(x_1,y_1),(x_2,y_2), ... \in \real \times \real, which is stable in the sense that for each interval AA \subseteq \real, (i) the limiting relative frequency of AA under x1,x2,...x_1, x_2, ... is governed by an unknown probability distribution μ\mu, and (ii) the limiting average of those yiy_i with xiAx_i \in A is governed by an unknown regression function m()m(\cdot). A computationally simple scheme for estimating m()m(\cdot) is exhibited, and is shown to be L2L_2 consistent for stable sequences {(xi,yi)}\{(x_i,y_i)\} such that {yi}\{y_i\} is bounded and there is a known upper bound for the variation of m()m(\cdot) on intervals of the form (i,i](-i,i], i1i \geq 1. Complementing this positive result, it is shown that there is no consistent estimation scheme for the family of stable sequences whose regression functions have finite variation, even under the restriction that xi[0,1]x_i \in [0,1] and yiy_i is binary-valued.

View on arXiv
Comments on this paper