An elementary analysis of ridge regression with random design

In this note, we provide an elementary analysis of the prediction error of ridge regression with random design. The proof is short and self-contained. In particular, it bypasses the use of Rudelson’s deviation inequality for covariance matrices, through a combination of exchangeability arguments, matrix perturbation and operator convexity