Abstract
For automating decision-making systems, we need models that learn from complex and dynamic macroeconomic factors. In this research, we propose a hybrid forecasting framework, Bayes-Sarimax. This method incorporates forecasts of dynamic time series probabili-ties. The aim is to improve credit risk assessment accuracy and increase explainability. The model includes linear dependencies between loan defaults and macroeconomic factors. These factors include inflation, general citizen income, salaries, rising GDP, and the relationship between private loans and GDP.The model was learned and validated using real macroeconomic data. This data origi-nated from a bank portfolio and encompassed both national and international sources. To rate macroeconomic stress due to loan defaults, we used lagged correlations, causal dependency, and impulse reactions. Bayesian layers improve the quality of models on limited data. They include a priori distributions and quantify uncertainty. Sarimax adds explainable time dynamics. Experimental results show the hybrid model outperforms classical SARIMAX and au-tonomous Bayesian models. It does better in both forecasting accuracy and clarity of decision support. This environment provides more reliable automation of cred-it-related decisions in intelligent control systems. It supports the development of adap-tive, data-driven financial control systems.