By José Azar (University of Navarra), Martin C. Schmalz (University of Michigan) & Isabel Tecu (Charles River Associates)
Dennis, Gerardi, and Schenone (2017) (DGS) claim to replicate the data construction and results of Azar, Schmalz, and Tecu (forthcoming) (AST). While their implementation of the main specifications in AST generates qualitatively similar results, they claim that AST’s baseline results are driven 1) by the use of passenger volume as regression weights and 2) largely by the top fifth percentile of markets in the passenger count distribution.
In this note, we show that these claims are factually incorrect. First, because DGS do not in fact replicate the data construction described in AST, their paper is of limited usefulness in showing the effect of deviations from AST’s empirical specifications. Second, we show that AST’s results are qualitatively robust to not weighting regressions. Third, AST’s results also hold on subsamples excluding the top fifth percentile of markets by passenger count. Additional evidence we present in this note suggests that DGS’s erroneous conclusions are driven by an incorrect treatment of ownership data as well as other differences in their sample’s characteristics compared to AST’s.
