Abstract

We develop and test a model of high quality annual report narratives. The model is trained and evaluated on reports published between 2007 and 2018 by London Stock Exchange-listed firms shortlisted by corporate reporting experts for an annual report award. We construct two sets of linguistic features to differentiate between shortlisted and non-shortlisted reports. The first set draws from extant accounting research and comprises reading ease (Flesch), net tone, forward-looking content, and hedging language. The second set is derived using semantic annotation and corpus linguistics methods to identify themes and language structures deemed by domain experts to support effective financial analysis and promote cognitive processing. In-sample tests reveal high-quality annual report narratives involve more discussion of strategy, more forward-looking commentary, and clearer presentation as measured by more relevancy markers, greater connectivity between narrative sections, simpler sentence structures, and fewer grammatical and negation words. Traditional bag-of-words proxies afford little explanatory power beyond corpus-derived features. Out-of-sample tests confirm predictive ability.