Cross-Validation Techniques for Resampling Housing Sales
Journal of Property Tax Assessment and Administration
This study examines internal model validation by extending the work of Jensen (2011) to verify the relevant assumptions upon which the model is formulated. This study provides residual diagnostics to check a housing valuation model for spatial and temporal correlations to reinforce that the model is methodologically sound. This study also examines external model validation by requiring double duty out of housing sales, using them to fit and cross validate the model. Cross-validation procedures, such as the k-fold, out-of-sample bootstrap, and standard leave-one-out techniques, are empirically shown to underestimate the variability of the percentage sales error by at least 30%. These common cross-validation techniques fail to follow valuation principles that are widely accepted by appraisers and tax assessors to value houses. This study introduces a block bootstrap technique that values resampled houses exactly as the model originally valued the subject property and, as a result, provides more realistic estimates of the variability of the percentage sales errors.
Department of Mathematics
Department of Economics
Original Publication Date
Ecker, Mark D. and Isakson, Hans R., "Cross-Validation Techniques for Resampling Housing Sales" (2022). Faculty Publications. 5333.