A law of the iterated logarithm for nonparametric regression function estimators (Q1074976)

Let \(m_ n(x)\) be a Nadaraya-Watson kernel type estimator or a Legendre orthogonal polynomial estimator of an unknown regression function m(x), based on an i.i.d. sample. A pointwise law of the iterated logarithm is proved for \(m_ n(x)-m(x).\) These results are parallel to those of \textit{P. Hall} [Z. Wahrscheinlichkeitstheor. Verw. Geb. 56, 47-61 (1981; Zbl 0443.62027)] on density estimation, and extend those of \textit{K. Noda} [Ann. Inst. Stat. Math. 28, 221-234 (1976; Zbl 0369.62068)] on the strong consistency of kernel regression estimators.