Overview

We have higher frequency data (monthly) from India. To leverage these data and show robustness to choice across methods, we adopt (i) an event study approach (using RDiT approach), (ii) forecast methods (using a time-series approach) and iii) placebo control approach using sales from prior periods. These approaches do not rely on the presence of a control group and therefore tradeoff higher frequency data for not fully accounting for temporal confounding factors.

In Table below, we present the results of the RDiT approach and forecast (SARIMA) methods. We use sales volumes from two prior time periods as placebo controls – i) July 2010 to June 2012, and ii) July 2009 to June 2011. We present the time periods we use as placebo controls in Table below. We employed generalized synthetic control method with data from the two placebo controls as control pools, and covariates (as we had covariate data for SKUs from India). Results from both the generalized synthetic control when using both the placebo controls are presented in Table below.

The results of (i) an event study approach (using RDiT approach), (ii) forecast methods (using a time-series approach) and iii) placebo control approach using sales from prior periods, are consistent with our main analysis results.