Brié, BjarneStouthuysen, KristofVerdonck, Tim2023-09-202023-09-202023http://hdl.handle.net/20.500.12127/7259We introduce a machine learning approach that accurately captures disclosure quality by quantifying the thematic content of climate reporting in annual reports. We then use our approach to analyze firms’ climate reporting practices in the context of the widespread European Non-Financial Reporting Directive. For a large sample of annual reports from 2010 to 2020, we find that firms significantly changed their climate reporting narratives in the periods following the announcement and implementation of the mandate. Using the Task Force on Climate-Related Financial Disclosures’ framework as the benchmark for high-quality climate reporting, we show that these changes correspond with improvements in disclosure quality. We further show that the comparability of reporting improves over time and provide first descriptive evidence of a more pronounced comparability change in the years following the implementation of the mandate. Overall, our results highlight the validity of our model and provide further descriptive evidence on the disclosure impact of nonfinancial reporting regulation. Our study also adds to the growing body of research applying machine learning to analyze information from annual reports.enCSR DisclosureClimate ChangeDisclosure RegulationMachine LearningTextual AnalysisAnnual Reports286361119751