A Match Made in Maastricht: Estimating The Treatment Effect of the Euro On Trade

Why do estimates of the European Monetary Union (EMU) effect on trade vary so greatly? Rose (2017) shows that the largest factor determining the size of EMU trade estimates is the choice of sample, with studies using only European or rich countries finding smaller impacts than those using more complete trade datasets. I push this question one step further, asking instead: what is the appropriate comparison group with which to study the euro’s trade impact? Using a first stage estimation of selection into the EMU and a robust propensity score weighting estimator, I extend the work of Millimet and Tchernis (2009) to a larger dataset of countries and years, showing that gravity estimates of the euro effect on trade are smaller when sample truncation and weighting brings the differences in observable characteristics between EMU and non-EMU pairs close to zero. Utilizing a Poisson pseudo-maximum likelihood approach, I find that estimates using this more robust estimator reflect the same pattern, but with significantly less initial upward bias. My work suggests that policy analysis in trade should be more careful to consider the comparability of “treated” and “control” observations, and more readily utilize propensity score methods as a data driven approach to rebalancing samples when differences across these groups are large.