Kellin Pelrine


Abstract

In this note we investigate the order of the unconditional bias of the Nadarya-Watson (Nadaraya, 1964; Watson, 1964) estimator for a multivariate regression. Surprisingly, previous attempts in establishing this result are either imprecise and technically deficient, or of limited use given the assumptions imposed (see inter alia, Glad (1998), Mack and Müller (1988), Pagan and Ullah (1999), and Scott (2015)). The results are also often conflicting (see inter alia, Choi et al. (2000), Chu and Marron (1991), Collomb (1981), and Glad (1998)). Unfortunately, our result here is incomplete, but we highlight the issues and suggest further ideas to resolve them.