With the use of artificial intelligence (AI) by investors set to grow, IR teams should track adoption by different funds and offer help sourcing relevant data, according to new research.
The transition remains at an early stage, however, with most asset managers not yet using AI techniques and fully ‘AI-powered’ funds accounting for just a small part of the overall buy side.
The study, produced by IHS Markit and due for release later today, investigates the impact of AI technology on both the asset management industry and tools designed for IR departments.
‘IROs should know which funds are moving in that direction, and when to start engaging to help them think through what datasets are going to help that evolution,’ says John Londono, global head of product, technology and innovation at IHS Markit.
‘If you inject yourself into that process of a firm evolving from traditional to more AI-powered, then you’ll always be part of the conversation.’
‘Gradually, they’re going to have machines making more and more decisions. But there will be, for the most part, still a human directing that evolution.’
For fund managers constantly searching for an edge, AI seems to offer great promise. Machine learning, a key subfield of AI, is far more adept than human researchers at finding patterns within large datasets.
Another technique, natural language processing (NLP), can be harnessed to pull insight from text sources at unprecedented speed. Increasingly, NLP is being focused on corporate information, such as press releases and earnings-call transcripts.
Yet a study last year by CFA Institute finds just 10 percent of portfolio managers had used AI techniques over the previous 12 months.
A different 2018 survey by BarclayHedge looked at adoption by the hedge fund industry, which unsurprisingly is at the forefront of efforts to embed AI technology into investment decision-making.
In this survey, more than half (56 percent) of respondents say they use a machine-learning approach in their investment process, a rise from 20 percent in the previous year.
Among early adopters of AI, the most common application is idea generation, says Londono. ‘What AI is very good at is sifting through reams and reams of data and trying to find, through quantitative methods, correlations,’ he says.
‘The AI models of today can surface these correlations that a human would probably not normally see, and that also might be a little bit difficult to understand.’
Some funds are going further and putting AI techniques and expertise at the heart of what they do, although IHS Markit’s study reveals this group of firms remains small in number and overall assets. Included in this group are larger firms like DE Shaw, Renaissance Technologies and Two Sigma, along with a string of smaller names.
‘There’s a handful of funds that you can argue are operating at some sense of scale with these techniques,’ says Londono. ‘Then there’s a longer tail of small upstart funds that aren’t managing that much money.’
The findings show that ‘truly AI-powered’ funds manage between 1 percent and 2 percent of total equity assets – a range that has not changed much over the last five years. In addition, many of these funds manage far less than $100 bn.
IHS Markit says there are several reasons why these funds may have struggled to grow in number or size, although there is no clear explanation. One of the main barriers could be the difficulty in explaining the decision-making of some AI models.
Large investors, like pension funds, may be unwilling to put money into AI funds ‘where they can’t, with confidence, get answers about why an investment decision was made,’ says Londono.
‘I think it’s a few years out where a machine will be able to tell an operator with confidence: this is why I made this choice. When that happens, I think you’ll probably start to see a lot more money flowing into these types of funds.’
Other reasons for a lack of investment, according to the report, may include difficulty in scaling up AI strategies and a reluctance to embrace technology that could see investment professionals lose their jobs.
The models operated by AI-led funds appear to have held up well during recent market volatility. The Eurekahedge AI Hedge Fund Index, which tracks 21 funds whose managers use AI and machine learning, was up 0.18 percent after the first four months of the year. Over the same period, the wider hedge fund industry was down 4.43 percent.
AI and IR
AI technology is also becoming available to IROs. Over recent years, AI-based products have launched covering areas like targeting, surveillance and, increasingly, sentiment analysis. Also referred to as opinion mining, sentiment analysis uses NLP to analyze text and identify the underlying feelings behind the words.
For some time, investors have used sentiment analysis to look at company announcements and earnings call transcripts to analyze the sentiment of management and how that may be changing. These tools are now being redesigned for IR teams, offering them the chance to understand the tone conveyed by their own communications.
As part of its research, IHS Markit took a close look at AI-based sentiment analysis tools created for IR teams. The report says there are solutions being marketed to IROs based on the idea that ‘subtleties in the language of the prepared statements and how management responds to questions could serve as leading indicators of market reaction’.
On the question of how effective these tools are for IR, the report says: ‘There is a fair amount of skepticism surrounding the ability of the NLP models to predict market impact from earnings calls. The primary reason is that it is not possible to draw a confident conclusion that one source of information would have such outsized influence on market reaction.’
The research notes that sentiment analysis tools could possibly ‘gauge a more generalized sentiment from the investor community’, although current applications have ‘shortcomings’ related to their datasets.
‘Unless you’re going to go out there and recruit a large body of sell-side and buy-side analysts to go through a huge corpus of prepared statements and then tag them, you’re going to have a hard time getting true signal out of something like that,’ says Londono.
‘Will that change over time? It probably will as these techniques get more sophisticated, or perhaps there might be a labor pool out there that’s willing to do something like that. But today, it’s not quite there yet.’
Financial communications has traditionally been a difficult area in which to ascertain sentiment due to the specific nature of the language. But providers of sentiment-analysis tools say effective analysis can be carried out using proprietary techniques and data sets.
Chris Ackerson, director of product search and AI at AlphaSense, says there is research suggesting alpha can be derived from earnings call sentiment. He adds, however, that his firm’s focus is more on helping investors understand the sentiment across large numbers of companies that individuals are not able to research manually.
His firm says it has built a training dataset over 10 years incorporating hundreds of thousands of labelled earnings call transcripts. The effectiveness of the AI model is determined by whether its assessment would match a human’s reaction to the text.
For IR teams, Ackerson says the interest in sentiment analysis is often based on understanding how other market participants, from quant funds to sell-side analysts, are interpreting the company's communications. There are some teams looking at whether this analysis can be used on documents prior to publication, he adds.