This paper proposes a new way to choose regression discontinuity models.
Candidate models are evaluated at placebo cutoffs away from the threshold, where the effect is assumed zero. The selected models have lower root mean squared error than the CCT approach in almost every simulation.
#econsky
