The strategy uses sentiment from WSJ photos and text to trade SPY, buying after negative sentiment spikes, holding two days, and allocating to Treasury bills otherwise.
ASSET CLASS: bonds, ETFs | REGION: Global | FREQUENCY:
Monthly | MARKET: bonds, equities | KEYWORD: Machine Learning, Newspaper
I. STRATEGY IN A NUTSHELL
The strategy trades S&P 500 (SPY) vs. 30-day Treasury bills using sentiment from WSJ articles and images (Sept 2008–Sept 2020). A transfer learning model (Google Inception retrained on DeepSent) classifies photos as positive or negative. PhotoPes and TextPes factors measure daily negative sentiment from images and text. If both exceed historical means, the strategy buys SPY on day 3 and sells two days later; otherwise, capital is held in risk-free assets.
II. ECONOMIC RATIONALE
Visual and textual sentiment affects investor expectations, with negative imagery causing short-term price movements lasting up to five days. This creates arbitrage opportunities by exploiting sentiment-driven pricing gaps. While implementation may incur higher costs in illiquid securities, out-of-sample results show positive certainty-equivalent returns, confirming the predictive value of combining text and image sentiment in equity trading
III. SOURCE PAPER
A Picture is Worth a Thousand Words: Measuring Investor Sentiment by Combining Machine Learning and Photos from News [Click to Open PDF]
Obaid, Khaled, Mississippi State University; Pukthuanthong, Kuntara, University of Missouri, Columbia
<Abstract>
By applying machine learning to the accurate and cost-effective classification of photos based on sentiment, we introduce a daily market-level investor sentiment index (Photo Pessimism) obtained from a large sample of news photos. Consistent with behavioral models, Photo Pessimism predicts market return reversals and trading volume. The relation is strongest among stocks with high limits to arbitrage and during periods of elevated fear. We examine whether Photo Pessimism and pessimism embedded in news text act as complements or substitutes for each other in predicting stock returns and find evidence that the two are substitutes
IV. BACKTEST PERFORMANCE
| Annualised Return | 15.06% |
| Volatility | 15.82% |
| Beta | N/A |
| Sharpe Ratio | 0.95 |
| Sortino Ratio | N/A |
| Maximum Drawdown | N/A |
| Win Rate | N/A |
