Mandatory CSR Reporting in Europe : A Textual Analysis of Firms’ Climate Disclosure Practices

We introduce a machine learning approach to quantify the thematic content of climate reporting in annual reports and use this to analyze firms’ climate reporting practices in the context of the widespread European Non-Financial Reporting Directive. We finetune ClimateBert to identify climate disclosure and fit a Structural Topic Model to capture the corresponding thematic content for a large collection of annual reports from 2010 to 2020. We find that firms significantly change their climate reporting narratives, corresponding with improvements in disclosure quality, and that firms start doing so before the implementation of the mandate. We further show that the comparability of reporting improves and provide first descriptive evidence of a more pronounced comparability change in the years following the implementation of the mandate. Taken together, our results highlight the validity of our model and provide further descriptive evidence on the disclosure impact of nonfinancial reporting regulation