Accurately predicting people’s movements within a small area is crucial for urban planning, transportation optimization, and emergency response preparedness. In this respect, emerging time series models tailored to handle complex seasonal patterns show promising potential. Identifying the most effective model for capturing traffic flows is of utmost importance for informed decision-making by policymakers. In this study, we compare the predictive performance of two approaches: a vector autoregressive model with dynamic harmonic components and the Facebook Prophet model. To this aim, we capture human mobility using data from the mobile phone network and analyze three traffic flow types (inflows, outflows, and internal flows) in Cellatica, a Municipality in the Province of Brescia. Employing a cross-validation strategy, we assess the models’ predictive ability using the MAPE. Our findings suggest that the multivariate model, which can capture the intricate correlation structure among various flow types, yields consistently better forecasts of traffic flows.