We use three supervised machine learning methods, namely linear discriminant analysis, quadratic discriminant analysis, and random forest, to predict corporate financial performance. We use a sample of 63 listed banks from eight emerging markets, covering 10 years from 2008 to 2017, using earning per share as a measure of performance. We use the design science research (DSR) framework to examine whether the textual contents of annual reports in previous years contain value-relevant information to predict future performance; thus, these contents can improve the accuracy and quality of predictive models. We combine two groups of variables in the proposed models. The first group is the sentiment analysis of disclosure tone in annual report narratives using the Loughran and McDonald dictionary (J Finance 66:35–65, 2011), while the second group is the quantitative properties of banks which consist of five variables, namely size, financial leverage, age, market-to-book ratio, and risk. Our analysis suggests that the random forest method provides the best predictive model. We also provide evidence on the accuracy and performance of predictive models that can be increased by incorporating disclosure tone variables as non-financial variables with financial variables. Interestingly, we find that the uncertainty variable is the most important disclosure tone variable. Finally, we find that size is the most important variable related to banks’ quantitative characteristics. Our study suggests that the analysis of tone through corporate narrative disclosures can be used as a complementary or diagnostic approach rather than an alternative in making decisions by different stakeholders such as analysts, investors, and auditors.

我们使用三种有监督的机器学习方法，即线性判别分析、二次判别分析和随机森林，来预测企业财务绩效。我们使用了来自 8 个新兴市场的 63 家上市银行的样本，涵盖了 2008 年至 2017 年的 10 年，使用每股收益作为绩效衡量标准。我们使用设计科学研究（DSR）框架来检查往年年度报告的文本内容是否包含与价值相关的信息来预测未来的表现；因此，这些内容可以提高预测模型的准确性和质量。我们在建议的模型中组合了两组变量。第一组是使用 Loughran 和 McDonald 词典（J Finance 66:35–65, 2011）对年度报告叙述中披露语气的情绪分析，而第二组是银行的量化属性，由规模、财务杠杆、年龄、市账率和风险五个变量组成。我们的分析表明随机森林方法提供了最好的预测模型。我们还提供了关于预测模型的准确性和性能的证据，可以通过将披露语气变量作为非财务变量与财务变量结合来提高预测模型的准确性和性能。有趣的是，我们发现不确定性变量是最重要的披露语气变量。最后，我们发现规模是与银行数量特征相关的最重要的变量。