Methods of adaptive demand forecasting in an unstable market environment based on the integration of macro- and microeconomic factors

Adaptive forecasting methods enable short-term forecasting of indicator dynamics, which is often crucial in dynamic and highly volatile economic environments. A comparative analysis of various adaptive approaches to demand forecasting is conducted, highlighting their limitations and advantages. A proprietary hybrid architecture for demand forecasting in an unstable market environment is proposed, based on the concept of a state space with hierarchical correction. The proposed architecture combines macroeconomic data and microeconomic indicators in a single state space, enabling the joint processing of disparate signals: macroeconomic shocks and operational microindicators, which in standard models typically require a choice. The statistical foundation is formed by a vector error correction model that captures stable long-term dependencies; nonlinear patterns are processed by LSTM networks and gradient boosting.