Analyzing The Dynamics of Financial Communication Language: “Measuring Financial Constraints using Embeddings” and “Measuring Sentiment of Non-GAAP Disclosures”.

Measuring Financial Constraints Using Embeddings: I propose and illustrate a novel method to capture corporate financial constraints. By “quantifying” textual information related to such constraints, my aim is to advance the use of textual analysis in measuring financial constraints directly from 10-K filings. This research offers a straightforward approach that is both timely and efficient. I employ a simple classification model and utilize a very powerful open-source language model. I use the Machine Learning text “Embeddings” principle by applying it to the “Liquidity and Capital Resources” subsection of the 10-K’s MD&A section. First, I find a distinction between equity financing constraints and debt financing constraints. Next, I find that the market responds more favorably to Debt financing compared to Equity financing. Further, I find that firm characteristics, such as firm size, age, and profitability are successful in explaining the dynamics for both equity and debt financing constraints.

Measuring Sentiment of Non-GAAP Disclosures: Non-GAAP financial reporting is ubiquitous in today’s corporate communications with investors. Many of these measures are not standardized and not directly comparable across companies. Although there is a large body of literature on non-GAAP reporting, no study has explicitly focused on the sentiment surrounding these disclosures. In this study, I propose a novel measure of sentiment of non-GAAP disclosures. To do so, I employ a FinBERT sentiment classification model using advanced language analysis. I find that, on average, sentiment around non-GAAP disclosures is slightly negative, suggesting that non-GAAP reporting is one channel managers can use to convey bad news to investors. It may also suggest that managers do not use non-GAAP information to opportunistically provide a better picture of the financial conditions of the company than what is actually occurring. Next, I find that sentiment of non-GAAP reporting has become more negative over time (while the prevalence of non-GAAP reporting has increased), with an interesting increase during the post-crisis years of 2009-2011. Importantly, I find that my constructed measure of non-GAAP sentiment varies significantly with firm characteristics: larger firms have a less negative sentiment in their communication of non-GAAP information with shareholders, and firms with more intangible assets communicate in non-GAAP terms with a more neutral or positive sentiment. Collectively, my results suggest that non-GAAP information is informative rather than misleading to investors and that sentiment around non-GAAP disclosures varies considerably across firms and over time.