HORIZON-CONDITIONED HYBRID FORECASTING: A DYNAMIC INTEGRATION FRAMEWORK BETWEEN ECONOMETRICS AND ARTIFICIAL INTELLIGENCE

Abstract

Economic forecasting increasingly operates at the intersection of econometric discipline and artificial intelligence driven flexibility. While traditional econometric models provide interpretability and theoretical consistency, they often struggle to capture nonlinear and structurally evolving dynamics. Conversely, AI-based forecasting models offer adaptive pattern recognition capabilities but raise concerns regarding interpretability and structural coherence. This study proposes a dynamic hybrid conceptual framework that integrates econometric and AI-based components within a forecast-horizon-conditioned structure. Unlike conventional hybrid models that rely on static residual modeling or simple forecast averaging, the proposed approach introduces dynamic weighting mechanisms that adjust according to forecast horizon, thereby structurally balancing linear and nonlinear components.

The framework preserves econometric interpretability in the short run while augmenting predictive flexibility in medium- and long-term horizons. To illustrate its internal logic and stability properties, a conceptual Monte Carlo-based validation is presented under stylized theoretical conditions. The study contributes to the forecasting literature by reframing hybridization as a structurally grounded integration process rather than a mechanical ensemble method, and by explicitly incorporating forecast horizon as a central determinant of model architecture.