Paras Anand, Fidelity International
Fidelity started using four AI tools (three already in practise, one still on trial) earlier this year, although building a team and creating adequate tools started eighteen months ago, Anand said.
The firm has hired data scientists to support the adoption of AI. Anand declined to estimate the costs involved, but said it was “modest”.
One of the firm’s tools is to detect “crowding and herding” behaviour.
“As an investor you are better off if you have a differentiated view of the market and go to less popular areas, but emotionally we are all drawn to the crowd.
“We built a tool which helps us get a comprehensive view of how popular a stock is.”
The software tool sends color-coded alerts when a stock is determined to be under-owned (green) or over-owned (red), he said.
A second tool aims to address confirmation bias – the inclination of investors to look for positive information when money has been allocated to a certain position.
Another AI software tool attempts to analyse the impact of loss aversion – investors’ bias toward avoiding losses rather than attempting to get equivalent gains.
A fourth tool seeks out “hidden information”. It has not yet been put into use, but Anand said it will aim to “identify any gaps between what might be happening at the company versus the shop window that is presented through their accounts”.
The story in the data
He said the software tools are not meant to generate out-performance, but are about providing marginal gains in investments.
“It is not because of these tools that your performance is going to improve by x amount. The benefit of these tools is that they point at things that are worth looking at.
“We try to find the story in the numbers.”
That said, the tools appear to have the potential for boosting returns.
According to Fidelity data, if the investment team had bought all the stocks that ended up in the “under-owned” green alert (according to the herding AI tool), the results would have been a 4% return, he said.
“Our statistical analysis showed that stocks globally that ended up in this [under-owned] category outperformed the MSCI All Country World Index by 400bps over the next two years.
“Had you sold most of the stocks in the crowded (over-owned) area in October, you would have had a very strong performance”.
Anand said AI is starting to show up as well in other parts of the investment process.
“For example, when we are sitting down with a company, we have a set of observations that come from the AI tools. That makes a change in the conversation with the CFO.”
He has also started to use them when he evaluates fund managers.
But there are limitations because AI is new, he admits. “They need an adoption time.”
Ultimately, Anand believes that AI won’t replace human intelligence in the market. “I don’t think an investment strategy solely driven by AI is going to be superior.”
AI is based on machine learning, which requires operating in a framework of well-structured problems. That is a crucial limitation, he said.
“When you think about the world of investment, you deal everyday with poorly-structured problems.”