Urban Land Valuation with Bundled Good and Land Residual Assumptions

Abstract

This paper develops a new approach to estimate the value of urban land. We extend AMM theory by adding the assumption of partial irreversibility. Bundled goods assumptions imply that land value with a structure can evolve differently than as-if vacant value, even in the first decades of structure life. We develop a hybrid model that nests bundled goods with land residual methods and we develop a new test of predictive accuracy. Granular house price indices produced by machine learning are used to estimate hybrid economic structure and land values.

We fit hybrid models to Maricopa County assessor data on houses up to 25 years old during a bust and recovery period (2007–2018). Results support structure value that evolves at some fraction (<1.0) of property value: i.e., partial bundling. The hybrid model achieves substantially lower Coefficients of Dispersion (CODs) than the land residual method, and both methods easily beat standard AVMs that do not separate land and structure values.

Keywords: Land valuation, property valuation, CAMA, machine learning

JEL Classification: R30, R31, R32