Abstract
Probabilistic machine learning models offer a distinct advantage over traditional deterministic approaches by quantifying both epistemic uncertainty (stemming from limited data or model knowledge) and aleatoric uncertainty (due to inherent randomness in the data), along with full distributional forecasts. These capacities are particularly appealing in light of growing regulatory concerns and the well-documented challenges of stability, interpretability, trustworthiness, accountability, and risk management in many machine learning applications. This review of probabilistic artificial intelligence in financial time-series uncertainty forecasting highlights several critical gaps in the field. These include a lack of standardized benchmarks and evaluation metrics, limited interdisciplinary collaboration, and insufficient financial interpretation of results. Collectively, these shortcomings hinder the ability to draw definitive conclusions about the performance of probabilistic models. The field remains nascent and fragmented, with most research published only recently and few studies building upon prior work, - likely due in part to the infrequent disclosure of code. We conclude that the potential for financial decision-making provided by probabilistic AI remains largely underutilized.