Suppose $\newcommand{\Bias}{\rm Bias}\hat\theta$ is an unbiased estimator of $\theta$ and $\operatorname{Var} \hat\theta =\lambda \theta^2$ for constant $\lambda$. Find the MSE of c$\theta$, where $c$ is a constant. From there, find the value of c that minimizes the mean squared error and deduce that $MSE_{c \hat\theta}=MSE_{\hat\theta}/1+\lambda$

So here's I've done so far. I think my main issue is confusing the c with the $\lambda$. I'm not so sure.

The formulas I'll be using are:

$MSE_{c \hat\theta}=\operatorname{Var} (c \hat\theta)+\Bias^2$ where $\Bias=c\theta-\theta$ hence:

$MSE_{c \hat\theta}=c\theta^{2}+((c-1)\theta)^2=(c^{2}-c+1)\theta$

To find a $c$ value that minimizes MSE, we just take the derivative at $c=0$ where we get $2c-1=0$ and $c=1/2$.

Now moving on to the fraction. I think $MSE_{ \hat\theta}=\operatorname{Var}_{\hat\theta} +\Bias^2=\lambda\theta^2$

From there we plug into the formula and get $(c^2-c+1)\theta^{2}=\lambda\theta^2/1+\lambda$, which I know isn't true.

What did I misinterpret and how can I change it?


If $\operatorname{Var} \hat\theta =\lambda \theta^2$ for constant $\lambda$, then wouldn't $\operatorname{Var} (c\hat\theta) =c^2\lambda \theta^2$ for constant $c$? Leading to $MSE_{c \hat\theta}=c^2\lambda\theta^{2}+((c-1)\theta)^2=(c^{2}(\lambda+1)-2c+1)\theta^2$. The derivative w.r.t. $c$ is thus $2\theta^2(c(\lambda+1)-1)$, which has a single root at $c=1/(\lambda+1)$. With that value of $c$, we have $MSE_{c \hat\theta}=(1/(\lambda+1))^2\lambda\theta^{2}+(((1/(\lambda+1))-1)\theta)^2=\lambda\theta^2/(\lambda+1)=MSE_{\hat\theta}/(\lambda+1)$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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