This study investigates the dynamics of automobile insurance loss metrics, emphasizing spatial patterns across different rating units (e.g., Forward Sortation Areas). By analyzing this spatial loss data, we reveal patterns and variations in auto insurance loss metrics over both spatial and temporal dimensions, offering essential insights for insurers and regulators aiming to refine risk assessment and pricing strategies. Using auto insurance loss data from Ontario, Canada, we identify pronounced differences in claim frequency, loss cost and claim severity between the sparsely populated northern regions and the densely populated south. Furthermore, our spatial analysis spans multiple accident half years, including those impacted by the COVID-19 pandemic, uncovering distinct temporal trends. By applying advanced spatio-temporal models, we enable loss predictions, equipping insurers to better navigate the evolving landscape of risk and uncertainty. This research advances our understanding of the dynamic nature of territorial risk within spatio-temporal contexts, providing valuable guidance for insurance companies and regulators in managing territory risk and its forecasting.