True structure change, spurious treatment effect? A novel approach to disentangle treatment effects from structure changes

This paper develops a new flexible approach to disentangle treatment effects from structure changes. It is shown that ignoring prior structure changes or endogenous regime switches in causal inferences will lead to false positive or false negative treatment effects estimations. A difference in difference in difference strategy and a novel approach based on Automatically Auxiliary Regressions (AARs) are designed to separately identify and estimate treatment effects, structure changes effects and endogenous regime switch effects. The new approach has several desirable features. First, it does not need instrument variables to handle endogeneities and it is easy to implement with hardly any technical barriers to the empirical researchers; second, it can be extended to isolate one treatment from other treatments when the outcome is the working of a series of treatments; third, it outperforms other popular competitors in small sample simulations and the biases caused by endogeneities vanish with sample size. The new method is illustrated then in a comparative study of supporting direct destruction theory on the impacts of Hanshin-Awaji earthquake and Schumpeterian creative destruction theory on the impacts of Wenchuan earthquake.