Artificial Intelligence Measurement of Disclosure (AIMD)

Abstract

Empirical research on voluntary disclosure lacks an appropriate measurement technique for quantifying the intensity of a firm's disclosure. In this paper, I introduce artificial intelligence measurement of disclosure (AIMD), a computerised technique for measuring disclosure using artificial intelligence, which derives disclosure proxies from English-language annual reports for 10 different information dimensions without human involvement. Criterion validity tests indicate that, controlling for a robust set of covariates and multiple statistical techniques, AIMD is negatively associated with information asymmetry as proxied by spreads and PIN. Furthermore, AIMD has construct validity when compared to the AIMR disclosure rating, Standard & Poor's Transparency and Disclosure Rating, several proprietary manual disclosure scorings and companies’ own assessment of their level of disclosure as indicated by a survey. I also demonstrate the applicability of AIMD as a cost-effective technique for measuring disclosure using a sample of 127,895 firm-year observations of companies regulated by the SEC.

Acknowledgements

I would like to thank Andrea Menini, the participants of the 2008 Financial Reporting & Business Communication (FRBC) Conference and the 2007 European Accounting Association (EAA) Annual Meeting. I am also grateful to two anonymous referees for their helpful comments, Christine Botosan for providing the raw AIMR data and Nick Ukiah for language editing.

Data Availability Statement

AIMD data is available from the author on request.

Notes

Studies claiming to measure the quality of disclosure use characteristic terms which are thought to indicate forward-looking, non-financial and non-quantitative information, for example.

Alternatively convergence can also be measured using confirmative factor analysis or structural equation analysis. These techniques are identical for the two variable problems examined here.

Although the rating covers around 200 companies, for most years results are publicly available for the top 50–110 companies only.

Some firms used US GAAP during a transitional period.

Financial institutions and insurance companies were excluded as they may use different wording and provide different content from manufacturing companies and service providers in their annual reports. Additional research is required into the heterogeneity or homogeneity of corporate documents with regard to particular sectors.

For instance, the abbreviation ‘EVA’ appears on the financial information N-gram list while the Christian name ‘Eva’ is part of the residual category which does not relate to corporate disclosure.