I agree. Synthetic controls can be powerful but they are not magic wand for causal analysis. The fit on the left seems too good to be true. It could be that it

I agree. Synthetic controls can be powerful but they are not magic wand for causal analysis. The fit on the left seems too good to be true. It could be that it using the outcome (Tesla's stock price) as part of the calibration, which biases your synthetic control. In ML speak it is an overfitted model. http://www.oekonometrie.uni-saarland.de/papers/SCM_Predictors.pdf http://www.oekonometrie.uni-saarland.de/papers/SCM_Predictor.... Also, in the package documention it says "The model also assumes that the relationship between covariates and treated time series, as established during the pre-period, remains stable throughout the post-period". In this case the covariate is the S&P500, so clearly the relationship didn't stay stable. As you mention, there was a dip in every tech company. This doesn't mean that Musk's tweet didn't affect the price, but this is no proof.