Van Lent asks how accountants can use their knowledge of measurement, that is, the mapping of transactions into financial/monetary terms, to learn more about firms’ activities and how firms impact society beyond their financial performance. Methodologically, the project will further refine its natural language processing algorithms by incorporating developments in the word embeddings literature as well as computational linguistics innovations that aim to detect “cheap talk” in written language.
How can accounting measurement principles be used to transform soft information in corporate disclosures into reliable metrics of the firm’s impact on society (and society’s impact on the firm)?
While corporate disclosures are a natural source of hard, quantitative information on corporations, these communications provide an immense amount of narrative information as well, which is often difficult to verify. This project examines such “soft information”. We investigate how to identify the sources and types of soft information, how soft information is disseminated by companies and financial intermediaries, how market participants process and respond to soft information, and what the consequences are of soft information in real and financial markets. We also develop natural language processing (NLP) algorithms that assist researchers in identifying soft information from (textual) sources, harvesting this information for large-scale data analysis, and turning this raw data into meaningful summary statistics in a replicable
We continue to structure the project into two complementary parts, substantive and methodological. In the substantive work packages, the central question will be how accountants can use their knowledge about measurement, i.e., the mapping of transactions into financial/monetary terms, to learn more about what goes on in firms and how firms impact society beyond their financial performance. Methodologically, we plan to further refine our algorithms by incorporating developments in the word embeddings literature as well as computational linguistics innovations that aim to detect “cheap talk” in written language. We also want to examine whether linguistic tools can be used to trace how shocks propagate in an economy.
Looking beyond the current funding period, the ambition of B10 is to enable citizens to use the data and methods developed in this project to improve society. Such “democratization of data” will help the ambition of showing that accountants can make the world a better place. Together, the proposed work will contribute significantly to the understanding of transparency.
The project has produced firm-level measures of risk and sentiment for US publicly listed firms. These measures are designed in particular to capture political risk, including specific topics such as health care, environment, and economic policy. The data are made publicly available on www.firmlevelrisk.com and the machine learning algorithm used to generate the measures is part of the replication package which is available on https://doi.org/10.7910/DVN/OBNRBP.