Lasso and the Methods of Causality

An exciting, quite new branch of econometrics studies how machine learning techniques can be adapted to consistently estimate causal effects. For example, the widely cited article by Belloni, Chernozhukov and Hansen (2014) introduces a post double selection method where one first runs two lasso regressions to select suitable control variables for a final OLS regression. When I first heard about the method, it had the alluring promise to provide an objective way to select control variables that may reduce the scope of p-hacking. But assumptions may be violated and there may be plausible settings where the method fails. Unfortunately, I lack the deep econometric background to have a good intuitive assessment of the assumptions. But luckily even without a deep grasp of asymptotic theory, one can always combine some intuition with Monte-Carlo simulation to assess an econometric method.