My understanding is that the Epanechnikov kernel is "efficient" in a mean squared error sense. Footnote 4 of Wikipedia's page defines the "efficiency" of a kernel as $$\sqrt{\int u^2 K(u)du} \int K(u)^2 du.$$

which is minimized by the Epanechnikov kernel. (I've seen this definition in places other than Wikipedia). I am struggling to map this definition directly into mean squared error.

For kernel density estimation of the density of $X$, the mean squared error I derived is,

$$\frac{1}{4} h^4 \left(∫_{-∞}^∞z^2 K(z)dz\right)^2 f^{''}(x)^2+\frac{1}{nh} f(x) \left(∫_{-∞}^∞ K(z)^2 dz\right)$$

For a local constant (Nadaraya-Watson) regression estimating $E[Y|X]$, my derivation of MSE results in: $$h^4 \left(\int u^2 K(u)du\right)^2 B^2 (x)+\frac{\sigma^2(x)(\int K(u)^2 du))}{nhf(x)} $$

Where $\sigma^2(x)$ is the variance of $Y$ at $X=x$.

Mean Integrated Squared Error would involve integrating over $x$.

Can anyone help me understand how the definition of "efficiency" as posted on Wikipedia is directly related to minimizing MSE? (or MISE/AMISE?) The components of Wikipedia's definition definitely appear in MSE, but it definitely seems distinct.



1 Answer 1


Kernels can be normalized based on $\int K(u)^2 du$. i.e. one could normalize such that $\int K(u)^2 du =1$, and the only term that affects efficiency is the so-called "roughness" of the kernel, $\int u^2 K(u)^2 du $.

Thus, mulitplying by $\int K(u)^2 du$ is just scaling based on the variance of the kernel (which could be normalized).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.