Stepwise
selection does not limit the extremeness of predictions, and
may even contribute to it by biasing the estimated regression
coefficients. The correct answer is (b), shrinkage methods are
intended to reduce the extremeness of predictions, and to make
predictions more reliable (=better calibrated).