Temporal Generalizability and Predictive Stability in Municipal Finance: Parallel Evidence from Norway and the United States

Abstract

This study evaluates temporal generalizability and predictive stability in municipal finance using NorwegianKOSTRA and U.S. Census government-finance data. Norway analyses combine discovery, false-discovery-ratecontrol, official municipality-continuity correction and held-out temporal validation; U.S. analyses use analogousfixed-effects models with influence and support diagnostics. A separate forecasting program uses rolling out-of-timeevaluation with a forecast-origin embargo.Three Norwegian associations passed discovery FDR control. Formal discovery-versus-validation tests sup-ported changes for unrestricted revenue per capita versus accumulated balance (BH q = 0.0058) and self-investmentfinancing versus debt burden (BH q = 0.0165), but not for home-service use (BH q = 0.732). Validation intervalsfor the first two include zero, and discovery selection may inflate initial effects, so these findings indicate limitedheld-out generalizability rather than proven structural breaks.In the United States, long-horizon interest-burden coefficients were strongly influenced by extreme ratio obser-vations. At h=2 the discovery coefficient fell from -22.98 to between -1.5 and -6.1 under every robustification. Asmaller period difference persisted under most support rules but changed form, from attenuation to sign reversal,and disappeared under trimming; at h=3 the period-change interpretation did not remain robust. Cross-countryrevenue-structure comparisons yielded no FDR-supported replication.Forecasting patterns also differed by country. Norway showed strong predictability relative to a pooled historicalmean, whereas U.S. RMSE rankings were weakly separated and dominated by extreme squared errors; MAErankings differed at longer horizons. Overall, the evidence supports cautious, horizon- and specification-dependentconclusions rather than universal fiscal relationships or a common forecasting architecture.