Ha. But this approach seems really problematic since it's based on the assumption that algorithms are good at predicting the counterfactual, while the reality is that the stock market is by nature extremely unpredictable.

Example 1: You could imagine the US gov announces a national quarantine, and Elon tweets this in response 1min later. The tweet didn't cause the stock drop, the quarantine did (a classic C caused both A and B problem). But this algorithm wouldn't know that.

Example 2: Imagine a court case ruling against Tesla is announced, and Elon immediately tweets. Same issue, but harder to detect. In the quarantine case you could assume Elon doesn't influence stocks besides Tesla, and observe that all of them are crashing.

Example 3: A really good earnings report for Tesla is announced, at the same time Musk tweets it's overpriced. The price cancels these out and stays steady. Now the algorithm gives a false negative, since actually the tweet prevented the price from going up.