2020 Volume 76 Issue 3 Pages 251-263
It is a well-known fact that the real estate market is segmented geographically. This phenomenon has attracted much research interest, and attempts have been made to extract regions where valuation standards are the same. Most studies presume a division structure according to specific geographical units. However, they might have failed to extract the actual condition of geographic segmentation because the real estate market has a hierarchical division structure, ranging from the municipality level to the neighborhood level.
This study proposes a new approach to identify geographical segmentation of the real estate market. We construct a real estate rent model with several regional explanatory variables that depend on different spatial resolutions and then implement the generalized fused lasso—a regression method for promoting sparsity—to extract regions where the valuation standard is the same. The proposed method is applied to the rent data of apartments in the Tokyo metropolitan area, and it confirmed the applicability of the proposed approach. The result indicates the characteristics of the real estate market in Tokyo: multi-scale segmentation occurs in the whole area, and neighborhood-level segmentation is particularly notable at the city center.