The max-MSE's of minimax estimators of variance in nonparametric regression (Q1894984)

Consider the nonparametric regression model \[ Y_i= g(t_i)+ \varepsilon_i, \qquad 1\leq i\leq n, \] where observations are taken at design points \(t_i\) for \(1\leq i\leq n\), and the errors \(\varepsilon_i\) are independent and identically distributed as normal distribution with mean zero and variance \(\sigma^2\). The response function \(g\) is assumed to belong to the space \[ W= \Biggl\{ g:\;g\text{ and } g' \text{ are absolutely continuous, and } \int_0^1 |g'' (t) |^2 dt< \infty \Biggr\}. \] Here, we deal with minimax estimators of \(\sigma^2\) defined by \textit{M. J. Buckley}, \textit{G. K. Eagleson} and \textit{B. W. Silverman} [Biometrika 75, No. 2, 189-199 (1988; Zbl 0639.62032)] based on the restricted class of the response functions \(W_C= \{g\in W\): \(\int_0^1 |g'' (t)^2 dt\leq C\}\).