Prediction intervals versus coefficient intervals

Why is uncertainty about a future return usually larger than uncertainty about its conditional mean?

A future observation includes irreducible outcome noise in addition to uncertainty about estimated model parameters. An interval for the conditional mean captures estimation uncertainty; a prediction interval must also account for the conditional outcome distribution. Heteroscedasticity, serial dependence and model misspecification complicate both. State the target, horizon and coverage assumptions rather than labelling every uncertainty band a prediction interval.