lukestein’s Twitter Archive—№ 1,621

@arpitrage @Andrew___Baker I totally buy your general point (really!) but not your reductio. “Weight 𝜆 on signal should be zero.” “I don’t know what ‘should’ means because I don’t know the objective function, but 𝜆=0.2 could be better at *prediction* than 0.” “But we get socially bad outcomes if 𝜆=1.”