I just ran two trillion regressions

The computational effort required to conduct a full model search to identify the most useful specification in problems that feature a large set of potential explanatory variables is widely perceived to be large. To circumvent or mitigate this challenge, the literature has proposed a host of techniques, many of which are not easy to implement. Using the example of a standard cross-country growth regression data set, we demonstrate that the computational effort in conducting a full model search will often be negligible. We provide an assessment of how this finding generalizes to model spaces of different sizes.