Forecasting Intermittent Demand for Luxury Goods Considering Zero-Inflated and Exogenous Shocks

Abstract

Traditional forecasting models often struggle with the intermittent demand, long-tail distribution, and zero-inflation inherent in luxury goods sales. To address this, we propose a three-stage hybrid ensemble forecasting model, RF-XGBoost-ZG. First, a Random Forest regressor captures primary time-series trends. Second, an XGBoost regressor fits the logarithmic residuals to capture nonlinear demand shocks from exogenous events. Finally, a Zero-value Gate classifier applies hard truncation and probabilistic smoothing to correct invalid demand. Empirical results demonstrate that the model significantly reduces extreme errors, balancing peak prediction accuracy with long-tail noise reduction. This framework provides a robust methodological tool for luxury retail inventory optimization and supply chain management.