Defining Geographic Markets Through Choice Sets: An Empirical Method Applied to Real Estate Brokerage

We introduce a choice-set approach to defining markets and a novel method to empirically recover geographic markets using machine learning, Spatial and Categorical Bayesian Clustering (SCBC). SCBC leverages the identity of the seller for each observation to capture market structures in a novel way that is not captured by purely distance-based methods. Applied to real estate agents in Stockholm (Sweden), SCBC classifies sales more accurately than the baseline K-means algorithm. Finally, we investigate the correct number of clusters and find that the optimal number of clusters
is close to the validation set based on industry knowledge.