Unveiling news impact on exchange rates: a hybrid model using NLP and LDA techniques

Purpose This study aims to investigate the influence of U.S. dollar-related news on EUR/US$ exchange rate using a novel hybrid news-fundamentals-based VAR model applied to 18 years of monthly data. Design/methodology/approach Leveraging Latent Dirichlet Allocation (LDA), the authors identify the top 5 U.S. dollar-related news topics, quantify the attention they receive over time using Shannon’s entropy, and integrate these news-generated metrics with news-constructed economic uncertainty indices and Taylor rule fundamentals into the VAR model. Through impulse-response analysis and forecast error decomposition, the authors examine how exchange rates react to shocks from the identified US$-related news topics and economic uncertainty captured by the news. Findings The findings reveal that news related to the US dollar and economic uncertainty account for 29% of long-term EUR/US$ variation. These results are robust, validated through robustness checks, Granger causality tests, sensitivity analysis and applying the same model to the GBP/USD exchange rate. Originality/value Combining news attention metrics with macroeconomic fundamentals enhances exchange rate identification, outperforming the models that rely solely on the Taylor rule or news variables.