Sustainability in property taxation: an application for revenue stability and fairness perspectives, the city of Buffalo, NY

The scholarly modeling of property tax has always posed a challenge, with two primary concerns to be addressed: first, maintaining sustainability in the collection, which is primarily a concern for local governments; and second, ensuring fair distribution, which is of greater concern for citizens. In today’s practices, assessment-based property tax increases unmatched expenses in bubble economies. There is a substitution problem in rapid falls, the tendency to not decrease the assessments gives way to black holes and opens the door to ghost cities. This paper proposes alternative approaches, aside from market/land value or last sold price, aimed at improving sustainability and fairness rates. The dataset examined is based on 93.7K records and 88 attributes for assessed value of properties within the City of Buffalo, the United States of America. Since the label (Total Value) is a numerical and continuous value, regression models are selected, where ensemble machine learning methods categorically work well with larger datasets, combined with weak learners, like decision trees. Stacked Ensemble led the least error for regression with 0.98 R2, followed by Gradient Boosting. Results show a 79% dominance of uncontrollable attributes, such as Land Value, Neighborhood, and Sale (last sold) Price, compared to controllable attributes, such as Total Living Area, Construction Grade, Second Story Area, and many others. This article suggests having a more balanced split between uncontrollable and controllable attributes would contribute to both sustainability and fair distribution.