Traders have long been using artificial intelligence to guide their actions. An interesting new approach was highlighted in a recent paper from the University of Chicago, which showed how methods taken from the autonomous transportation sector could be used in trading.
The researchers explain that traders have frequently used various price charts to help them predict possible investment returns. These charts provide a range of visual shorthand patterns to help, such as “double top” and “head and shoulders”. They have tapped into the kind of pattern-recognition methods used in self-driving vehicles to also help investors spot these patterns.
Visual representation
The paper underpins the notion that there is often extremely valuable, not to mention actionable, information contained in stock charts and the visual representations of data they provide.
“We have heard that traders can translate price charts into signals,” the researchers explain. “Our findings suggest that the shape of the data contains enough information for trading. This justifies what technical traders do.”
The researchers tapped into data from past prices, and also the momentum they were going in, each of which was expressed graphically. They found that these graphical representations can be extremely useful in guiding trading strategies, especially in the short term. The effect wanes over time, however, as the fundamentals become more important.
As such, the researchers believe that training a machine learning system to read charts can provide higher returns, with fewer risks, than more traditional methods of predicting returns. The approach used by the researchers builds on convolutional neural network technology, which is the same kind of technology used by autonomous vehicle technology to accurately identify pedestrians and road furniture. It’s an approach the researchers believe could become a valuable investment aid for traders to help them become less reliant on hunches and intuition.
Trading support
“A technical trader uses prior knowledge to define patterns,” they explain. “A CNN has no prior knowledge. Without using any existing chart, I am going to ask the CNN to learn from the price curve to extract useful information for prediction.”
Importantly, the study suggests that the approach can be valuable even if its utilizing relatively small or even incomplete data sets. Once the system has been trained on markets with ample data, they believe it can be deployed in markets with less plentiful data available.
Indeed, this method of transfer learning could even be used to help traders to analyze new and emerging asset classes if they share some fundamental features with more mature markets.
