Bi-Directional Hierarchical Recurrent Neural Network for Inflation Forecasting (BiHRNN)
Bi-Directional Hierarchical Recurrent Neural Network for Inflation Forecasting (BiHRNN)
Maya Vilenko, M.Sc. student at the department of Industrial Advisor: Dr. Noam Koenigstein
Abstract:
Inflation prediction is essential for guiding decisions on interest rates, investments, and wages, as well as for enabling central banks to establish effective monetary policies to ensure economic stability. The complexity of predicting inflation arises from the interplay of numerous dynamic factors and the hierarchical structure of the Consumer Price Index (CPI), which organizes goods and services into categories and subcategories to capture their contributions to overall inflation. This hierarchical nature demands advanced modeling techniques to achieve accurate forecasts. Traditional approaches to predictive modeling with hierarchical data include training separate models for each series within the hierarchy or using a single model for the combined data. While the former can mitigate underfitting by focusing on specific data segments, it often leads to overfitting due to limited data for individual models. The latter leverages larger datasets but is computationally expensive and struggles to address the varying dynamics across hierarchical levels. We introduce BiHRNN, a novel modeling approach that strikes a balance between these extremes by leveraging the hierarchical structure of datasets like the CPI. BiHRNN facilitates bidirectional information flow between hierarchical levels, where higher-level nodes influence lower-level ones and vice versa. This is achieved using hierarchical informative priors applied to the parameters of recurrent neural networks (RNNs), enhancing predictive accuracy across all hierarchy levels. By integrating hierarchical relationships without the inefficiencies of a unified model, BiHRNN offers an effective solution for accurate and scalable inflation forecasting.
Bio:
Maya Vilenko is a Master's student in the Department of Industrial Engineering at Tel Aviv University. She currently works as a Data Scientist at PayPal. Maya holds a Bachelor of Science in Industrial Engineering from Tel Aviv University.
