Language of Bankruptcy: Analysis of Word Sequences and Word Context

Abstract

Academics and practitioners searched for reliable indicators of companies’ failure focusing only on quantitative data such as financial ratios and market variables. However, recent literature aims to quantify textual information of financial reports studying features such as topics and words’ co-occurrences, confirming their usefulness in predicting company bankruptcy. In this work, we propose a new approach to analysing texts that focuses on sentences interpreted as ordered sequences of words. We propose a new approach, based on Language Model, to predict the company’s bankruptcy that was released in the next year. Given the high predictive power of the model, we investigate the sentences of texts to gain insights into how failing companies’ language differs from the nonfailing one. Our approach allows us to move away from fixed word-lists, exploring linguistic features to understand how a word is used in different contexts. The results of our analysis lead us to observe that the concept of bankruptcy can take on different meanings arising from the different legitimisation strategies that companies facing bankruptcy may use.