Diagonal correction for a quadratic density derivative estimate

ID: diagonal-correction-for-a-quadratic-density-derivative-estimate

Let be an independent and identically distributed sample from probability density function , take and , and write . Assume the kernel for density estimation has square-integrable second derivative. Define
This U-statistic obeys whenever differentiation and the displayed integrals are justified. Under the usual convolution regularity this is .
The exact relation to the second derivative kernel density estimator is
Indeed, expand the square into ordered pairs; its diagonal terms are all . Both subtracting this diagonal and changing the normalization from to are needed for the displayed U-statistic. This removes the diagonal sampling contribution, while retaining smoothing bias of an estimator relative to . The corrected value can be negative in a finite sample even though its target is nonnegative.

New to topics? Read the docs here!