Advancing BiHRNN: Enhanced Bi-Directional Hierarchical Recurrent Neural Network for Inflation Forecasting
Advancing BiHRNN: Enhanced Bi-Directional Hierarchical Recurrent Neural Network for Inflation Forecasting
Chen Hershko, M.Sc. student at School of Industrial & Intelligent Systems Engineering
Advisor:Prof. Noam Koenigstein
Abstract:
Accurate inflation forecasting is essential for informing monetary policy, guiding investment strategies, and promoting economic stability. This thesis builds upon the original Bi-directional Hierarchical Recurrent Neural Network (BiHRNN) model, transforming it into a robust and scalable framework for disaggregated inflation forecasting. BiHRNN ia an extension of the Hierarchical Recurrent Neural Network (HRNN) which processes hierarchical time series by aggregating bottom-up signals, introducing the concept of bidirectional information flow between hierarchical levels. In this work, we significantly enhance the BiHRNN architecture to better exploit the structure of the Harmonised Index of Consumer Prices (HICP), enabling richer interactions across hierarchy levels through structured parameter sharing and improved training procedures. These methodological advances allow for more accurate modeling of cross-level dependencies without incurring the computational costs associated with unified, flat models. We conduct an extensive empirical evaluation using Eurostat’s HICP dataset, enriched with a broad set of economic indicators. The results demonstrate that the enhanced BiHRNN consistently and significantly outperforms both HRNN and the original BiHRNN baseline across multiple levels of the hierarchy. By turning a conceptual prototype into a thoroughly validated solution, this work establishes BiHRNN as a state-of-the-art approach for hierarchical inflation forecasting.
Bio:
Chen Hershko is an M.Sc. student at the department of Industrial Engineering at Tel Aviv University, specializing in Data Science. Chen holds a B.Sc. degree in Industrial Engineering from Ben-Gurion University. Her research, supervised by Dr. Noam Koenigstein, focuses on improving inflation forecasting using a novel model (BiHRNN) that leverages the hierarchical structure of price data (HICP) for more accurate and detailed Bio: Chen Hershko is an M.Sc. student at the department of Industrial Engineering at Tel Aviv University, specializing in Data Science. Chen holds a B.Sc. degree in Industrial Engineering from Ben-Gurion University. Her research, supervised by Dr. Noam Koenigstein, focuses on improving inflation forecasting using a novel model (BiHRNN) that leverages the hierarchical structure of price data (HICP) for more accurate and detailed predictions across European countries.
