Risky news and credit market sentiment

Abstract

The nonlinear nexus between financial conditions indicators and the conditional distribution of GDP growth has recently been challenged. We show how one can use textual economic news combined with a shallow Neural Network to construct an alternative financial indicator based on word embeddings. By design the index associates growth-at-risk to news about credit, leverage and funding, and we document that the proposed indicator is particularly informative about the lower left tail of the GDP distribution and delivers significantly better out-of-sample density forecasts than commonly used alternatives. Speaking to theories on endogenous information choice and credit-market sentiment we further document that the news-based index likely carries information about beliefs rather than fundamentals.