To Understand Financial Disclosures, Transform Words Into Numbers
If there is one realm which is Big Data by definition, it is Accounting. Yet even in this area, numbers alone are not enough to explain complex business phenomena and the study of the textual content of the financial disclosures is also necessary. The related research of Francesco Grossetti, Professor at the Bocconi Department of Accounting and Fellow of the Bocconi Institute for Data Science and Analytics, is contained in the working paper "Crafting Financial Disclosure. Does Tone Suffice?", which focuses precisely on the textual analysis of financial statements and their impact on corporate performance. It is co-signed by colleagues Miles Gietzmann and Craig M. Lewis at Vanderbilt University in Nashville, Tennessee.
"In the study we apply machine learning algorithms to analyze the contents of the MD&A (Management's Discussion & Analysis) sections present in the annual reports which US public companies must submit to the Securities and Exchange Commission (SEC), the supervisory body enforcing the law against market manipulation", explains the researcher. "These are rather long, discursive paragraphs, which are sometimes difficult to interpret and that are usually studied by observing certain metrics such as the recurring terms, the tone of voice, or the presence of particular sentences.
Instead, here we have trained a series of Naive Bayes algorithms to classify each sentence along five different dimensions: tone; optimism; specificity and quality of the speech; whether it is frank and direct; or evasive and passive-aggressive. Each of these dimensions was scored on a scale from 1 to 5." The study covered 118,000 MD&A sections of as many financial documents released between 2001 and 2018, which is equivalent to about 30 million sentences that were analyzed by software to understand the word content and how they were perceived by the investors who read them.
"A text can be interpreted from different points of view. Mathematically speaking it means that there are, in theory, infinite combinations and interpretations of words, and this is a problem from an empirical point of view," summarizes Grossetti. "Here we try instead to better explain which elements investors pay attention to. It may seem trivial, but it is not because this issue has a direct consequence on the actions taken by investors and therefore on companies' individual destinies. We posit an investor that is rational, capable of interpreting almost every signal and maybe even able to understand if, for example, an MD&A is more forced or less forced, and identify hidden pitfalls. A positive tone does not necessarily equate with positive meaning. Conversely, a negative tone can have favorable implications if the sentence as a whole is optimistic and looks toward the company's future. Basically, there are not only signals of market sentiment, but also other dimensions that need to be evaluated to have a clearer understanding of how financial communication is constructed and what effects it can have".