An innovative procedure for smoothing parameter selection

Smoothing with penalized splines calls for an automatic method to select the size of the penalty parameter λ. We propose a not well known smoothing parameter selection procedure: the L-curve method. AIC and (generalized) cross validation represent the most common choices in this kind of problems even if they indicate light smoothing when the data represent a smooth trend plus correlated noise. In those cases the L-curve is a computationally efficient alternative and robust alternative.