Improving the prediction of firm performance using nonfinancial disclosures: a machine learning approach

Purpose - The purpose of this study is to test whether the prediction of firm performance can be enhanced by incorporating nonfinancial disclosures, such as narrative disclosure tone and corporate governance indicators, into financial predictive models. Design/methodology/approach - Three predictive models are developed, each with a different set of predictors. This study utilises two machine learning techniques, random forest and stochastic gradient boosting, for prediction via the three models. The data are collected from a sample of 1,250 annual reports of 125 nonfinancial firms in Pakistan for the period 2011–2020. Findings - Our results indicate that both narrative disclosure tone and corporate governance indicators significantly add to the accuracy of financial predictive models of firm performance. Practical implications - Our results offer implications for the restoration of investor confidence in the highly uncertain Pakistani market by establishing nonfinancial disclosures as reliable predictors of future firm performance. Accordingly, they encourage investors to pay more attention to these disclosures while making investment decisions. In addition, they urge regulators to promote and strengthen the reporting of such nonfinancial information. Originality/value - This study addresses the neglect of nonfinancial disclosures in the prediction of firm performance and the scarcity of corporate governance literature relevant to the use of machine learning techniques.