Date of Award

Summer 7-16-2021

Degree Type


Degree Name

Doctor of Philosophy (PhD)




Lee, Yoonseok


average treatment effects, Causal inference, Model averaging, subsampling, treatment effects

Subject Categories

Economics | Social and Behavioral Sciences


This dissertation contains essays on causal inference with model averaging. The first essay presents a theoretical derivation of a model-averaging-based average treatment effect estimator. The second essay provides comparison of predictability of treated counterfactual outcome between model averaging and other methods. The third essay is an empirical study evaluating the economic impact of Ukraine's 2013 conflict.

The first essay constructs a new average treatment effect estimator based on model averaging in a panel data setting. The estimator is shown to be asymptotically unbiased and consistent. Its asymptotic distribution is derived, which turns out to be non-normal and non- standard. A subsampling procedure is then applied to obtain valid inference. Simulation results show that the proposed estimator compares favorably with alternative estimators in out- of-sample prediction accuracy under a common factor structure.

The second essay further compares predictability of treated counterfactual outcome between model averaging and other methods under more general set-ups. The simulations show that the model averaging and penalized regression methods yield more accurate counterfactual prediction than the model selection methods. We also find evidences that if the predictors (e.g., control units' outcomes) are more correlated, the model averaging methods have more accurate prediction than the penalized regression, and vice versa.

The third essay evaluates the economic impact of Ukraine's 2013 conflict using a comparative case study. A modified synthetic control method is applied to account for potential spillover from the conflict on Ukraine's neighbouring countries. The results show that Ukraine's real GDP was reduced by 29.7% from late-2013 to the end of 2015. The spillover effects are detected in every quarter since the conflict began. Furthermore, negative spillover effects are found in countries selected by the modified synthetic control.


Open Access

Included in

Economics Commons