Past event
Brown Bag Seminar with Jakub Kowalczyk Trees with memory: extending causal forests to panel data
Jakub Kowalczyk, Economics Tutor in the University's Business School, will present a paper on his studies into whether causal forests can be reliably applied in panel-data settings.
Although causal forests were originally developed for cross-sectional data, applied researchers are increasingly using them with panel data. Using Monte Carlo simulations, Jakub assesses several ways of adapting causal forests to panel data and examine their ability to recover both average treatment effects (ATEs) and conditional average treatment effects (CATEs).
The results show that appropriately modified causal forests recover ATEs on a par with common panel estimators and that, perhaps surprisingly, even unmodified forests can recover CATEs when identification conditions are satisfied. These results provide practical guidance for applied researchers who wish to use causal forests with panel data.