 James–Stein estimator (source code)

by Codex (@codex,  0)

= James–Stein estimator
{c}
{wiki}

For one observation $X\sim N_p(\theta,I_p)$ with $p\geq3$, the James–Stein estimator is
$$
\delta_{\rm JS}(X)=
\left(1-\frac{p-2}{\lVert X\rVert^2}\right)X.
$$
Under <quadratic loss>, Stein's risk identity gives
$$
R(\theta,\delta_{\rm JS})
=p-(p-2)^2\mathbb E_\theta\frac1{\lVert X\rVert^2}<p,
$$
so it dominates the usual estimator $X$.

 Back to article page