Abstract We propose a novel measure of fiscal policy divergence based on public budget data. First, we quantify policy similarity across municipalities based on the cosine similarity of their budget allocations, showing strong correlations with geographic proximity and differences in socio-demographic characteristics. Next, we predict budget similarity out-of-sample using a flexible machine-learning model trained on pairwise differences in local characteristics, defining fiscal divergence as the municipality-year average absolute prediction error. In an empirical application to 8000 Italian municipalities (2000–2015), we show that electoral-cycle patterns are concentrated on the expenditure side of the budget: expenditure divergence is lower in pre-electoral years than in election years, whereas revenue divergence follows a different and less stable pattern.
Bello et al. (2026) studied this question.