## Interview Question

## Interview Answer

3 Answers

▲

0

▼

Basically same as DW: ( delta(0) + delta(1) )/2

▲

0

▼

It is very hard than that to prove the results for given the choice in all probability distributions.

you must find the density f(x) such that:

f is maximising Var(f) = int{0->1} x^2f(x) dx - (int{0->1} xf(x) dx)^2

under the constraint int{0->1} f(x) dx = 1

writing optimality conditions you get that f must be infinite in 0 and 1 and you can use the symmetry argument to get the DW result.

## Add Answers or Comments

To comment on this, Sign In or Sign Up.

You'd want the values to be at either 0 or 1, since anything in between just brings them closer to the mean (not fully rigorous, but should be obvious enough). let p be the density at 0, so (1-p) is the density at 1. Then mean = (1-p). Variance is: p*(1-p)^2+(1-p)*p^2 = p-2p^2+p^3+p^2-p^3 = p-p^2

We want to maximize variance, to take the first derivative and set it to 0:

1-2p = 0

p=1/2

Hence, half the density is at 0, half at 1.

There may be a more elegant argument.