A Picture is Worth a Thousand Words: Measuring Investor Sentiment by Combining Machine Learning and Photos from News
65 Pages Posted: 10 Dec 2018 Last revised: 16 Jan 2020
Date Written: January 15, 2020
By applying machine learning to accurately and cost effectively classify photos based on sentiment, we introduce a daily market-level investor sentiment index (Photo Pessimism) from a large sample of news photos. Between 1926 and 2018, Photo Pessimism predicts market return reversal and increase in trading volume. The return predictability pattern is concentrated among stocks with high limits to arbitrage and during high uncertainty and low investor distraction periods. Photo Pessimism complements sentiment in news text. Moreover, Photo Pessimism has over five times stronger predictive power when we focus only on business news. These results are consistent with behavioral models, but inconsistent with theories of media content as a proxy for new fundamental information.
Keywords: Investor Sentiment, Behavioral Finance, Return Predictability, Machine Learning, Big Data
JEL Classification: C53, G10, G17
Suggested Citation: Suggested Citation