We propose a Bayesian nonparametric approach for assessing macroeconomic tail risk using a time-dependent Dirichlet process mixture model. The model is applied to a datasets that covers several decades across both OECD and non-OECD countries, and is shown to captures fluctuations in extremely negative outcomes in consumption level changes. The analysis reveals left-skewed distributions consistent with the association of downside risks, and allows to evaluate these aspects dynamically and in comparison to upside risks. Such insights are invaluable for policymakers and investors, offering nuanced perspectives to navigate shifting economic landscapes and implement tailored risk management strategies effectively.