Is There a Gold Standard or Need for a City-Centric Approach for Sales Tax Revenue Forecasting

Abstract

The accuracy of sales tax revenue forecasting is essential for local governments, as they rely on these forecasts to develop their annual budgets. Previous research has focused on identifying gold-standard forecasting methods with high average accuracy across multiple cities. However, such approaches may still produce inaccurate predictions for specific municipalities, making this scholarship less relevant to practitioners. Our research addresses the gap in the existing literature by focusing on the relative accuracy of forecasts from the municipal perspective rather than the overall average accuracy across all municipalities—a city-centric approach—to identify variations in various machine learning and traditional revenue forecasting methods. Here, we show that following the steps of PREE: (P) prepare, (R) run multiple models, (E) evaluate against benchmarks, and (E) evaluate overall performance can help to maximize the accuracy of sales tax revenue forecasting at the municipal level. The high variability in model performance across municipalities highlights the risks associated with relying on a single gold-standard forecasting approach. Instead, practitioners should focus on forecasting processes, such as PREE.