I'm an economics student who's trying to dive into causality. I've tried looking into intervention analysis, but I can't seem to find many sources on forecasting using historical data as an approach.

I'm trying to infer causality based on an intervention event.

E.g. I own a store, and run a promotion for 2 weeks.

Can I build a forecast using historical sales data and use it as a counterfactual to measure the impact of my promotion (compared forecasted sales vs actual observed sales)? What are the flaws in this approach?

  • $\begingroup$ I do this all the time, very straight forward and simple and easily explainable. I don’t see any major drawbacks. You can never infer causality unless it’s a randomized experiment. $\endgroup$ – forecaster Jul 25 '18 at 17:26
  • $\begingroup$ It would seem to me that a problem might be that there are many models consistent with your historical data that yield different predictions for the counterfactual future data under no promotion. Your causal effect estimate would depend highly on the model used. $\endgroup$ – Noah Jul 25 '18 at 18:19
  • $\begingroup$ @forecaster, so this approach will allow me to quantify the impact of the intervention but I won't be able to state whether it was causal? E.g. Store sales received a 5% lift, but the case is out whether it was due to the promotion or not. $\endgroup$ – bellwether Jul 25 '18 at 19:04

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.