Delineating Market Areas Used for Mass Valuation Using Geographically Weighted Regression (GWR) and Hierarchical Cluster Analysis (HCA)

This paper aimed to develop a straightforward method for government property tax assessment offices to automate baseline market area delineation using geographically weighted regression (GWR) and hierarchical clustering analysis (HCA). The methodology initially integrates local R2 values from a GWR hedonic model/automated valuation model (AVM), as well as the individual latitude coordinate and longitude coordinate of each sale, into an HCA algorithm. To test the useability of the identified clusters as market areas in a mass valuation context, a hedonic model that considered each property’s physical features, time of sale, and proposed cluster, was constructed. The coefficient for each cluster dummy variable was statistically significant and the model yielded predictions that were in line with accuracy and uniformity thresholds in the IAAO’s Standard on Ratio Studies, both overall (city-wide) and for each market area proposed through clustering. Our findings are particularly relevant to government property tax assessors, who, with market areas that are current and well-defined, can improve office efficiencies and yield more accurate and equitable tax assessments. The study has a number of practical and policy outcomes. The findings will help more assessment offices realize the benefits of regression-based AVMs and reduce typical lag time associated with implementing regression-based valuation models, particularly when existing assessment neighborhood classifications are too granular and numerous to serve as dummy variables (a very common state of most assessment offices) – or are simply outdated – and location is otherwise unable to be directly represented in the model.