Estimators: Square in expected value VS Variance?

28 Views Asked by At

I've spent so much time trying to proof to myself these are equivalent, but it's not working out.

Given some $X = u + Y$,

where $Y \sim N(0,1)$.

I have some estimator $Z = 2x_0 + \sum_{i=0}^N x_i^2 + 2x_{N+1}$

$Var[z] = E[Z^2] = E[(2x_0 + \sum_{i=0}^N x_i^2 + 2x_{N+1})^2] = 4 + N + 4 = N + 8$

because the variance is distributed.

$ Var[2x_0] + Var[\sum_{i=0}^N x_i^2] + Var[2x_{N+1}] $

However, when I try to distribute the terms as a polynomial instead, I definitely don't get that answer. I know I'm doing something wrong somewhere.

$Var[z] = E[Z^2] = E[(2x_0 + \sum_{i=0}^N x_i^2 + 2x_{N+1})^2] = E[4x_0^2 + 2x_0\sum_{i=0}^N x_i^2+...+2x_N\sum_{i=0}^N x_i^2+4x_{N+1}^2]$

1

There are 1 best solutions below

6
On BEST ANSWER

Your calculations of the variance are not correct.

You can write $X\sim N(u,1)$. And $X^2 = u^2 + Y^2 + 2uY$. Thus, $$E[X] = u$$ and $$E[X^2] = u^2+1,$$ this is just the second moment of $X$.

Then the variance of $Z$ is defined as \begin{aligned} \text{Var}[Z] &= E[ (Z - E[Z])^2 ]\\ &= E[Z^2] - (E[Z])^2 \end{aligned}

Note that $$ E[Z] = 4u + (N+1)(u^2+1). $$