Forecasting weekly inflation in China with bottom-up, top-down, and combined frameworks

This study systematically compares bottom-up, top-down, and hybrid forecasting frameworks for high-frequency (weekly) inflation, leveraging China’s internet-based Consumer Price Index (iCPI). The bottom-up approach utilizes eight disaggregated price components, while the top-down framework incorporates 10 macro indicators. Empirical analyses across multiple models, including the traditional univariate autoregressive (AR) benchmark, Principal Component Analysis (PCA), three-pass regression filter (3PRF), and its Markov-switching extension (MS-3PRF), demonstrate that integrating new information or adopting alternative frameworks can significantly improve inflation forecasting. Notably, the MS-3PRF, designed to accommodate structural breaks, delivers substantial gains over various horizons, and the combination of bottom-up and top-down frameworks yields additional improvements. While no single model univgersally dominates, the MS-3PRF emerges as the most robust. These findings challenge the conventional dominance of AR models in inflation forecasting and provide the first systematic evidence for China’s high-frequency context. By highlighting the predictive value of disaggregated data and hybrid forecasting frameworks, this study offers actionable insights for real-time inflation monitoring and policymaking amid economic volatility.