Can train/test split help standard econometrics?

Imagine we are presented with the following data-set:

medium.com