Statistics · Model fitting · Medium
A mean pulled towards zero
A sample average x̄ is an unbiased estimate of a mean μ and has a standard deviation of 1. In truth μ = 2. You estimate μ by c × x̄, where c is a fixed number. Which value of c gives the smallest mean squared error?
a fraction, an expression or a decimal to two significant figures · 5/8 · ln 2 · pi · inf
Answer and solution
Answer
4/5 ≈ 0.8
Solution
Shrinking trades a little bias for less variance; the best factor is μ² / (μ² + variance).
MSE(c) = c² × 1 + (1 − c)² × 4. The derivative 2c − 8(1 − c) is zero at c = 8/10 = 0.8, where the MSE is 0.64 + 0.16 = 0.8, below the 1 of the unshrunk average.
Model fitting comes up in 6 firms' roundsTwo Sigma · Tower Research · Millennium · +3
These 8 rounds ask this kind of problem live, with follow-ups. Every round on every track is in Premium.