The second moment method (2nd MoM) is a popular technique to prove that a random variable X of interest is positive with high probability. It has been widely used to show the existence of specific structures in random graphs, to show the existence of solutions for random constraint satisfaction problems etc. In some situations, the 2nd MoM fails despite X being positive with high probability. In this talk, we examine the failure of the vanilla 2nd MoM in the context of analyzing the compression performance of Sparse Regression codes. Building on recent work by Coja-Oghlan and Zdeberova, we present a refinement of the 2nd MoM and show that sparse regression codes with optimal encoding achieve the Gaussian rate-distortion function for any ergodic source.