We build a corpus of over 51/2 million news articles on 20 large US firms over the 10-year period from January 2001 to December 2010, and use it to study the time-varying nature of the relation between media-expressed firm-specific tone and firm-level returns. By estimating a series of separate rolling window vector autoregressive (VAR) models for each firm, we show how media-expressed negative tone impacts firm-level returns episodically in ways that vary across firms and over time. We find that firms experience prolonged periods during which media-expressed tone has no effect on returns, and occasional episodes when it has a significant impact. During the significant episodes, its impacts are sometimes quickly reversed and at other times they endure - implying that media comment and analysis can sometimes be sentiment (or noise), but it can also contain value-relevant information or news. Our findings are in general consistent with efficiently functioning markets in which the media assists with the processing of complex information.
- Market efficiency
- Media-expressed tone, negative sentiment
- Textual analysis