A0844
Title: A signature synthetic control method for unbalanced panels
Authors: Pietro Emilio Spini - University of Bristol (United Kingdom) [presenting]
Stefan Hubner - University of Bristol (United Kingdom)
Abstract: The aim is to generalize the synthetic control methods to unbalanced panels where the overlap in the time index sets is very limited or absent altogether. Empirically relevant cases include control units with missing data and systematic differences in sample periods and/or frequencies. The approach relies on signatures, a nonparametric feature extraction method. After extracting the signatures of treatment and potential control units in the pre-intervention period, a synthetic control is constructed directly on the signatures. It is shown that the generalized method collapses to standard synthetic controls on a balanced panel. The properties of the resulting synthetic-control estimator are studied for the average treatment effect on the treated, and its convergence rates under standard conditions on the nonparametric component, which is treated as a nuisance function.
