‘When Is a Liability Not a Liability’ Revisited: Should Financial Text Analysis Abandon General Lexicons?

38 Pages Posted: 25 Jan 2015

See all articles by Brent Kitchens

Brent Kitchens

University of Virginia - McIntire School of Commerce

Debanjan Mitra

University of Connecticut - Department of Marketing

Joseph Johnson

University of Miami - Department of Marketing

Praveen Pathak

University of Florida

Date Written: July 1, 2014

Abstract

Finance scholars increasingly believe automated text analysis using a general lexicon misclassifies the tone of financial terms and is ineffective at predicting firm returns. We show that research originating this belief failed to use required word-sense disambiguation, leading to inaccurate results and conclusions regarding the effectiveness of general lexicons. Contrary to this research, we find that a disambiguated general lexicon accurately identifies tone in financial documents. Moreover, disambiguated general lexicons provide advantages of generalizability, precision, and objectivity. Our results imply that general lexicons should not be abandoned in financial text analysis, and disambiguation is critical for this type of analysis.

Keywords: word sense disambiguation, sentiment analysis, financial text analysis

Suggested Citation

Kitchens, Brent and Mitra, Debanjan and Johnson, Joseph and Pathak, Praveen, ‘When Is a Liability Not a Liability’ Revisited: Should Financial Text Analysis Abandon General Lexicons? (July 1, 2014). Available at SSRN: https://ssrn.com/abstract=2554656 or http://dx.doi.org/10.2139/ssrn.2554656

Brent Kitchens (Contact Author)

University of Virginia - McIntire School of Commerce ( email )

P.O. Box 400173
Charlottesville, VA 22904-4173
United States

Debanjan Mitra

University of Connecticut - Department of Marketing ( email )

Storrs, CT 06269
United States

Joseph Johnson

University of Miami - Department of Marketing ( email )

United States

Praveen Pathak

University of Florida ( email )

PO Box 117165, 201 Stuzin Hall
Gainesville, FL 32610-0496
United States

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
96
Abstract Views
892
Rank
492,631
PlumX Metrics