Topic-Level Narrative Novelty, Accruals Quality, and Market Liquidity: Evidence from German Annual Reports

Corporate annual reports cover economically distinct topics, yet existing measures of year-over-year narrative change typically aggregate these disclosures into a single value. We examine whether the associations between narrative change and firm outcomes depend on which topic changes. Using transformer-based language embeddings, we measure the year-over-year semantic novelty of more than 5,000 German annual reports from 2007 through 2019 and decompose it into four disclosure domains: financial reporting, operations, capital and tax, and governance, control, and regulatory disclosure. Greater financial-reporting novelty is associated with lower absolute abnormal accruals across three accrual models. A one-standard-deviation difference corresponds to approximately 9 percent of the sample mean. Capital-and-tax novelty has opposite-signed conditional coefficients that differ significantly from the financial-reporting coefficients in two of the three models, although the positive coefficients are not individually statistically significant. The composite novelty measure is negatively associated with absolute abnormal accruals and, more tentatively, with lower Amihud illiquidity, though not with the proportion of zero-return days. The decomposition localizes this liquidity association within the governance, control, and regulatory domain, and the component analysis suggests it is concentrated in governance content. These findings show that topic-level decomposition can reveal distinct and potentially offsetting associations that an aggregate textual measure obscures.