A Picture Is Worth a Thousand Words: Market Sentiment From Photos
59 Pages Posted: 10 Dec 2018 Last revised: 15 Dec 2018
Date Written: November 18, 2018
This paper proposes a new market-level investor sentiment index (Photo Sentiment) constructed from photos in the financial press. We apply a machine learning technique to classify a large sample of photos in The Economist based on perceived sentiment. Photo Sentiment is measured as the average probability the photos have negative sentiment minus the average probability the photos have positive sentiment. Between 1997 and 2017, Photo Sentiment is found to predict short-term return reversal, trading volume and market volatility, and explains flows between equity and money market funds. A trading strategy based on Photo Sentiment outperforms the market index by 2.6% on an annual basis while assuming less risk.
Keywords: Investor Sentiment, Behavioral Finance, Return Predictability, Machine Learning, Big Data
JEL Classification: C53, G10, G17
Suggested Citation: Suggested Citation