Statistics · Regression · Medium
A regressor read with error
The true relation is y = 1.5x + noise, where x has variance 4 and the noise is independent of x. You cannot observe x itself, only w = x + error, where the error has variance 1 and is independent of both x and the noise. With a very large sample you regress y on w, with an intercept. What slope do you get?
a fraction, an expression or a decimal to two significant figures · 5/8 · ln 2 · pi · inf
Answer and solution
Answer
6/5 ≈ 1.2
Solution
Noise in a regressor inflates the denominator only, so the slope shrinks towards zero by the factor var(x) / (var(x) + var(error)).
cov(y, w) = 1.5 × var(x) = 6. var(w) = 4 + 1 = 5. Slope = 6 / 5 = 1.2: the true 1.5 times 4/5.
Regression comes up in 11 firms' roundsHudson River Trading · Goldman Sachs · D. E. Shaw · +8
These 16 rounds ask this kind of problem live, with follow-ups. Every round on every track is in Premium.