Investors can be a moody bunch, and their moods can impact returns and investment performance, so what if those moods could be predicted?
New research from RMIT University has shown a picture really is worth 1,000 words, with a newly developed algorithm predicting investor moods using Getty Images.
Early seeds of this idea came from the University of Missouri, where researchers devised a way to index daily investor sentiments using news story photos.
RMIT lead author on the new research Dr Angel Zhong built on this idea by training the algorithm to analyse top lists of Getty’s editorial section. Through machine learning, the algorithm produces a score that can predict whether the market will have a good or bad day.
How does the algorithm work?
Dr Zhong told Yahoo Finance the algorithm gave a score of 0 to 0.92 to determine how pessimistic a photo was.
“Using machine learning, artificial intelligence … extracts the mood - the sentiment of pessimism conveyed in the photo - and then we use that measure to predict stock returns, and returns and trading bodies around the world in 37 countries,” she said.
The research team from RMIT University and Swinburne University of Technology have been able to harness the algorithm to increase global accuracy of the results, published in Finance Research Letters.
What would be considered a ‘good’ or ‘bad’ day photo?
A sunny field with bright colours is likely a good image, while a scene depicting the burnt out remains of a bushfire would likely be a bad image.
One of the ‘bad’ days of photos from the research is from April 16, 2003 during the Iraq War. The day was dominated by images of soldiers, war scenes and tense press conferences. This highly negative mood was scored at around 0.73 out of a maximum of 0.92. The global stock market index reflected this in returns of -0.41 per cent.
Other days with notable bad scores might be months like December, or events like September 11, 2001 or days of significant natural disasters.
“The algorithm will analyse photos, they look at things like the colour, facial expression, body movements and the object in a photo,” Dr Zhong said.
‘Bad day’ lasts for 2 days of returns
Researchers were surprised to uncover that a ‘bad day’ photo impacted market returns for around two days, an insight that could have implications for traders.
“If you are saying that today, while the market is negative, so, of course it is typical that the stock return will drop, nothing surprising there, but it did not for the next day as well,” Dr Zhong said.
“That means it is predictable, and you can implement a trading strategy based on that.”
Language no barrier
This area of research continues to evolve as technology improves. Previously researchers would use historical stock market data, and then textual analysis to make similar predictions.
The problem with textual analysis, says Zhong, is that it was primarily on English-language news, which failed to capture many of the leading global financial markets. Using the algorithm “transcends language barriers” and is able to be successful across so many global financial markets.
Future to grow the algorithm
Dr Zhong's team hopes to build on this research and develop a country-specific algorithm to align better with individual markets.
“I’m interested in developing more new measures to simplify this process and to demystify how industry behavioural biases affect...financial markets.”