Can a model that fits observed demand well still recover causal price response, substitution, and counterfactual outcomes when prices and promotions are endogenous?
CausalDemand pairs synthetic retail scanner panels with marketing-copy product descriptions that carry the true substitution geometry. Demand is simulated from a known data-generating process; in half the cells, promotion depth responds to a hidden demand shock, so estimators that ignore endogeneity fit… See the full description on the dataset page:
https://huggingface.co/datasets/jean-jsj/CausalDemand.
