European Central Bank working paper develops translation-free daily newspaper sentiment indices for euro area forecasting

The European Central Bank published Working Paper No 3122 by Manuel Medina Magro and Lorena Saiz presenting a dictionary-based method to derive daily sentiment indicators from euro area economic newspaper articles without translating the text, and reporting that the resulting measures can support nowcasting and short-horizon forecasting of output, labour market conditions and inflation. Using around 3 million Factiva-sourced articles from major newspapers in Germany, France, Italy and Spain, the authors construct topic indices for Economic News Sentiment (ENS), Labour News Sentiment (LNS) and Inflation News Sentiment (INS) based on the co-occurrence of topic keywords and directional-change terms within a narrow word window, alongside two recession proxies: an R-word index (share of articles mentioning “recession” or “crisis”) and an R-risk index (share of recession-related articles also mentioning “risk” or “uncertainty”). Aggregated Big-4 country measures co-move strongly with standard survey indicators and GDP, and in expanding-window VAR exercises the news-based indicators deliver lower one-quarter-ahead GDP forecast errors than common benchmarks in the 2004–2019 sample, with recession-word measures performing best in downturns. A daily probit model using 90-day moving averages of ENS and its squared term produces out-of-sample recession probabilities with an area under the ROC curve of 0.93 for the sovereign debt and COVID-19 recessions, while the term spread provides little incremental value. For inflation, INS is reported to track HICP inflation and household inflation expectations and to improve inflation forecasting during the 2020–2023 high-inflation period, including in VAR specifications that use INS and a quadratic term.