Forward-looking statements in corporate social responsibility (CSR) reports and firms’ future CSR performance

Purpose This study aims to examine the predictive value of forward-looking statements (FLS) in corporate social responsibility (CSR) reports issued by US public firms. Specifically, the authors assess whether the disclosure level and linguistic features of FLS in CSR reports contain information about future CSR performance. The predictions rest on two contrasting conceptual perspectives, namely, informative disclosure and impression management. Design/methodology/approach The authors analyze CSR reports and CSR performance data for US public firms over the period 2000–2019. Fine-tuning the Robustly Optimized BERT Pretraining Approach (RoBERTa) language model, the authors introduce FLS-RoBERTa and Financial-RoBERTa, transformer-based deep neural network models that identify FLS and analyze their tone, respectively. Findings The authors find that greater FLS quantity and a more positive FLS tone are associated with lower subsequent CSR performance, whereas higher-quality FLS – less boilerplate and more specific – are associated with higher subsequent performance. These results persist when all four attributes are included jointly, indicating that FLS quantity, quality and tone provide distinct, incremental information. Overall, the findings suggest that greater quantity, lower quality and more optimistic tone are consistent with impression management, whereas higher-quality FLS provide useful information about future CSR performance. Practical implications Practitioners should interpret FLS in CSR reports with caution. Regulators may consider establishing guidelines to standardize such disclosures. Researchers can use the introduced language models to enhance the analysis of corporate disclosures. Originality/value To the best of the authors’ knowledge, this study is among the first to systematically investigate FLS in CSR reports as predictors of future CSR performance. Additionally, it introduces new language models for analyzing the content of disclosures, offering methodological contributions to the literature.