News-Based Indices on Country Fundamentals

We propose a novel method to extract information on macro fundamentals based on regular expressions and compare its performance vis-a-vis two popular alternatives in the current finance literature based on predefined dictionaries and supervised learning techniques. We create news indices about fundamentals according to these techniques on a news dataset by Reuters and compare how the techniques fare at (i) capturing observed economic surprises (macro announcements compared to Bloomberg survey expectations) and (ii) correctly predicting labels on a manually classified test sample of the news text corpus. We demonstrate that our methodology is better able to identify and discriminate among fundamentals than the alternative techniques. We further show the benefit of our fundamental news indices in an econometric application that investigates the fundamental content of asset prices in the sovereign credit risk arena.