Automatic Summarization of Corporate Disclosures

Abstract

Managers often provide summaries of key disclosures, such as earnings releases and MD&A. Manager-provided summaries may, however, be prone to bias in tone and content. In contrast, automatic, algorithm-based summaries have the potential to provide useful summary information with less bias than management summaries. This paper investigates characteristics of management and automatic summaries for corporate disclosures and the effect of summaries on investors’ judgments. Specifically, we conduct three experiments to investigate how automatic summaries compare to management summaries on several dimensions (e.g., usefulness, bias), and how summaries affect investor information processing, beliefs about firm fundamentals (e.g., performance, risk), and valuation judgments. Our results suggest that automatic summaries compare favorably to management summaries for earnings releases, but fare less well for MD&A. Further, investors who receive an earnings release accompanied by an automatic summary arrive at lower valuation judgments, and are more confident in those judgments, compared to investors who receive the same earnings release with a management summary. Our findings provide input to recent discussions by policy makers on the use of summaries for corporate disclosures.