Predicting Market Inflation Expectations with News Topics and Sentiment

This study presents a novel approach to incorporating news topics and their associated sentiment into predictions of breakeven inflation rate (BEIR) movements for eight countries with mature bond markets. We calibrate five classes machine learning models including narrative-based features each country, find that they generally outperform corresponding benchmarks do not include such features. Logistic Regression XGBoost classifiers deliver the best performance across countries. complement these results feature importance analysis, showing economic financial are key drivers in our predictions, additional contributions from related health government. examine cross-country spillover effects narrative on BEIR via Graphical Granger Causality confirm existence US Germany, while other considered only influenced by local narrative.