High-Frequency Tourism Flow Forecasting in a Historical City Center: A Rolling Origin Evaluation of DHR, Splines, and TBATS Models

Introduction: Tourism flows are critical for urban planning and destination management. This study analyzes daily visitor dynamics within the historical center of Brescia, Italy, leveraging high-resolution mobile network data. Predicting arrivals at a daily frequency introduces significant analytical challenges due to complex, overlapping seasonal patterns.

Methods: To model daily fluctuations, a robust time-series forecasting framework was applied to mobile network tracking indicators. Three statistical approaches were compared: Dynamic Harmonic Regression (DHR), Splines, and TBATS models. These estimators were evaluated using a strict Rolling Origin procedure across 120 randomly selected working and non-working days over a one-year horizon to guarantee out-of-sample validity. The empirical application isolated daily inflows from three strategic European markets: France, Germany, and the United Kingdom.

Results: Predictive accuracy was assessed by separating the dataset into working days, non-working days, and a combined global horizon. Forecasting performance was evaluated using a diverse set of error metrics, placing specific emphasis on the scale-free measures that establish a direct comparison against a naïve benchmark.

Conclusions: Integrating high-frequency mobile network data with advanced statistical modeling significantly enhances localized tourism monitoring. Ultimately, these insights support actionable, data-driven decisions for short-term crowd management and sustainable urban planning, establishing a scalable methodological benchmark for historical city centers.