A0686
Title: Sub-model aggregation for scalable spatially varying coefficient modeling
Authors: Daisuke Murakami - The Institute of Statistical Mathematics (Japan) [presenting]
Shonosuke Sugasawa - Keio University (Japan)
Abstract: The aim is to develop an approach that aggregates/combines global and local sub-models to build a flexible and scalable spatial regression model, including a spatially varying coefficient model. To aggregate sub-models, a generalized product-of-experts method is used, which is widely used in machine learning literature. The aggregated spatial regression model has the following properties: (i) computationally efficient; (ii) each sub-model can be estimated independently; (iii) the marginal likelihood is available in closed form. Owing to (ii), the proposed method is capable of modelling complex spatial patterns. Owing to (iii), its model accuracy is easily compared across specifications. The accuracy and computational efficiency of the proposed method are compared with conventional methods through Monte Carlo simulation experiments. Then the method is then applied to an analysis of residential land prices in Japan.