AI-Analysis of CEOs’ Letters
An AI-based analysis of German CEOs’ letters to the shareholders
Does it make a difference to investors and analysts how CEOs communicate? The HHL Chair of Accounting and Auditing investigated this question in a collaborative study with Kirchhoff Consult and Precire Technologies. For this study, the authors analyzed the psycholinguistic characteristics of CEO letters by using artificial intelligence (AI).
“Managers will have to get used to the fact that what they say is tracked not only by financial analysts but also by Artificial Intelligencies.”
The analysis shows:
- According to AI interpretations, CEO letters of German DAX companies tend to be overly aggressive and impulsive, and subpar innovative,
- Management and corporate governance factors leave their marks on the tonality of CEO letters – e.g., dividend policy, shareholder structure.
- The tonality of CEO letters is empirically associated with the uncertainty of financial analysts’ forecasts – e.g., dispersion of earnings forecasts.
Managers’ tonality influences interpretation of hard facts
Prof. Dr. Henning Zülch, Chairholder of Accounting and Auditing, recognizes the potential of using AI in capital market research:
“Digitization pushes the limits of the feasible and the measureable. This also holds true for reporting and investor relations. Managers will have to get used to the fact that what they say is tracked not only by financial analysts but also by Artificial Intelligencies – with as of yet unforeseeable consequences. However, the use of AI also helps us to analyze and understand our own communicative behavior. How does our counterpart’s tone and charisma influence our interpretation of seemingly hard facts? The current study raises this question in an impressive way.”
For investors and financial analysts alike, the study’s results provide notable insights into the language of CEO letters, its determinants and influence on analysts’ forecast dispersion.
You can download the complete study (in German) directly from our website: