Asymptotically Unbiased Synthetic Control Methods by Density Matching

Masahiro Kato and Akari Ohda

Journal of Causal Inference.

In one sentence. Standard synthetic control weights are biased because donor outcomes are correlated with the error term; matching distributions rather than fitting observed outcomes yields an asymptotically unbiased estimator and a full counterfactual distribution.

Where the bias comes from

Synthetic control methods predict the counterfactual outcome of a treated unit as a weighted combination of untreated units, with weights chosen to match the pre-treatment outcome path. The difficulty is an endogeneity problem: the outcomes of the untreated units used as regressors are correlated with the error term, so weights fitted to observed outcomes inherit a bias that does not vanish with more pre-treatment periods.

Matching distributions instead

The paper assumes that the outcome distribution of the treated unit can be approximated by a weighted mixture of the untreated units' distributions, and estimates the weights by matching that mixture to the treated unit's distribution rather than by fitting realized outcomes. Three consequences follow:

Experiments confirm the behaviour implied by the theory.

Related

See the full publication list.